doi
stringlengths 10
10
| chunk-id
stringlengths 1
4
| chunk
stringlengths 1
1.66k
| id
stringlengths 10
10
| title
stringlengths 19
148
| summary
stringlengths 345
1.92k
| source
stringlengths 31
31
| authors
sequence | categories
sequence | comment
stringlengths 4
284
⌀ | journal_ref
stringclasses 14
values | primary_category
stringclasses 16
values | published
stringlengths 8
8
| updated
stringlengths 8
8
| references
list |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2301.11305 | 32 | In contrast, zero-shot methods generalize relatively easily
to new languages and domains; DetectGPT’s performance
in particular is mostly unaffected by the change in language
from English to German.
While our experiments have shown that DetectGPT is effective on a variety of domains and models, it is natural to
wonder if it is effective for the largest publicly-available
LMs. Therefore, we also evaluate multiple zero-shot and supervised methods on two 175B parameter models, OpenAI’s
GPT-3 and AI21 Labs’ Jurassic-2 Jumbo. Because neither
API provides access to the complete conditional distribution
4The overall ease of detecting machine-generated fake writing
corroborates anecdotal reporting that machine-generated creative
writing tends to be noticeably generic, and therefore relatively easy
to detect (Roose & Newton, 2022).
6
Zero-Shot Machine-Generated Text Detection using Probability Curvature
0.00 0.08 0.16 0.04 0.12 0.24 0.20
Fraction of GPT-J-generated news article re-written0.60.70.80.91.0Detection AUROC
Rank
DetectGPTLogRank
LikelihoodEntropy
Figure 5. We simulate human edits to machine-generated text by | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 33 | Rank
DetectGPTLogRank
LikelihoodEntropy
Figure 5. We simulate human edits to machine-generated text by
replacing varying fractions of model samples with T5-3B generated text (masking out random five word spans until r% of text is
masked to simulate human edits to machine-generated text). The
four top-performing methods all generally degrade in performance
with heavier revision, but DetectGPT is consistently most accurate.
Experiment is conducted on the XSum dataset.
for each token, we cannot compare to the rank, log rank, and
entropy-based prior methods. We sample 150 examples5
from the PubMedQA, XSum, and WritingPrompts datasets
and compare the two pre-trained RoBERTa-based detector
models with DetectGPT and the probability thresholding
baseline. We show in Table 2 that DetectGPT can provide
detection competitive with or better than the stronger of the
two supervised models, and it again greatly outperforms
probability thresholding on average.
5.2. Variants of Machine-Generated Text Detection
Detecting paraphrased machine-generated text. In practice, humans may manually edit or refine machine-generated
text rather than blindly use a model’s generations for their
task of interest. We therefore conduct an experiment to | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 34 | text rather than blindly use a model’s generations for their
task of interest. We therefore conduct an experiment to
simulate the detection problem for model samples that have
been increasingly heavily revised. We simulate human revision by replacing 5 word spans of the text with samples
from T5-3B until r% of the text has been replaced, and
report performance as rvaries. Figure 5 shows that DetectGPT maintains detection AUROC above 0.8 even when
nearly a quarter of the text in model samples has been replaced. Unsurprisingly, almost all methods show a gradual
degradation in performance as the sample is more heavily
revised. The entropy baseline shows surprisingly robust
performance in this setting (althought it is least accurate
on average), even slightly improving detection performance
up to 24% replacement. DetectGPT shows the strongest
detection performance for all revision levels.
Impact of alternative decoding strategies on detection.
While Table 1 suggests that DetectGPT is effective for
5We reduce the number of evaluation samples from 500 in our
main experiments to reduce the API costs of these experiments.XSum SQuAD WritingPrompts
Method top- ptop-ktop-ptop-ktop-p top-k
logp(x) 0.92 0.87 0.89 0.85 0.98 0.96 | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 35 | logp(x) 0.92 0.87 0.89 0.85 0.98 0.96
Rank 0.76 0.76 0.81 0.80 0.84 0.83
LogRank 0.93* 0.90* 0.92* 0.90* 0.98 0.97
Entropy 0.53 0.55 0.54 0.56 0.32 0.35
DetectGPT 0.98 0.98 0.94 0.93 0.98 0.97
Table 3. AUROC for zero-shot methods averaged across the five
models in Table 1 for both top- kand top-psampling, with k=
40andp= 0:96. Both settings enable slightly more accurate
detection, and DetectGPT consistently provides the best detection
performance. See Appendix Tables 4 and 5 for complete results.
detecting machine-generated text, prior work notes that
the decoding strategy (i.e., temperature sampling, top- k,
nucleus/top- p) can impact the difficulty of detection. We repeat the analysis from Section 5.1 using top- ksampling and
nucleus sampling. Top- ksampling truncates the sampling
distribution to only the khighest-probability next tokens; | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 36 | nucleus sampling. Top- ksampling truncates the sampling
distribution to only the khighest-probability next tokens;
nucleus sampling samples from only the smallest set of tokens whose combined probability exceeds p. The results
are summarized in Table 3; Appendix Tables 4 and 5 show
complete results. We use k= 40 , andp= 0:96, in line with
prior work (Ippolito et al., 2020). We find that both top- k
and nucleus sampling make detection easier, on average.
Averaging across domains, DetectGPT provides the clearest
signal for zero-shot detection.
Detection when the source model is unknown. While
our experiments have focused on the white-box setting
for machine-generated text detection, in this section, we
GPT-J GPT-Neo GPT-2Scoring ModelGPT-J GPT-Neo GPT-2Base Model0.92
(0.02)0.83
(0.04)0.79
(0.02)
0.64
(0.06)0.97
(0.01)0.83
(0.02)
0.60
(0.09)0.85
(0.05)0.99
(0.00)0.85
0.81 | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 37 | (0.02)
0.60
(0.09)0.85
(0.05)0.99
(0.00)0.85
0.81
0.81
0.72 0.88 0.87
Figure 6. DetectGPT performs
best when scoring samples
with the same model that generated them (diagonal), but
the column means suggest that
some models (GPT-Neo, GPT2) may be better ‘scorers’ than
others (GPT-J). White values
show mean (standard error)
AUROC over XSum, SQuAD,
and WritingPrompts; black
shows row/column mean.explore the effect of using
a different model to score a
candidate passage (and perturbed texts) than the model
that generated the passage.
In other words, we aim
to classify between humangenerated text and text from
model A, but without access to model Ato compute log probabilities. Instead, we use log probabilities computed by a surrogate model B. We consider three models, GPT-J,
GPT-Neo-2.7, and GPT-2,
evaluating all possible combinations of source model
and surrogate model (9 total). We average the performance across 200 samples
from XSum, SQuAD, and
7 | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 38 | evaluating all possible combinations of source model
and surrogate model (9 total). We average the performance across 200 samples
from XSum, SQuAD, and
7
Zero-Shot Machine-Generated Text Detection using Probability Curvature
60M 220M 770M 2.7B0.50.60.70.80.91.0Detection AUROC
5 perturbations
60M 220M 770M 2.7B
25 perturbations
Random
GPT2-sm
GPT2-md
GPT2-lg
GPT2-xl
Mask filling model size (# parameters)
Figure 7. There is a clear association between capacity of maskfilling model and detection performance, across source model
scales. Random mask filling (uniform sampling from mask filling
model vocabulary) performs poorly, reinforcing the idea that the
perturbation function should produce samples on the data manifold.
Curves show AUROC scores on 200 SQuAD contexts.
WritingPrompts. The results are presented in Figure 6,
showing that when the surrogate model is different from the
source model, detection performance is reduced, indicating
that DetectGPT is most suited to the white-box setting. Yet
we also observe that if we fix the model used for scoring
and average across source models whose generations are | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 39 | that DetectGPT is most suited to the white-box setting. Yet
we also observe that if we fix the model used for scoring
and average across source models whose generations are
detected (average within column), there is significant variation in AUROC; GPT-2 and GPT-Neo-2.7 seem to be better
‘scorers’ than GPT-J. These variations in cross-model scoring performance suggest ensembling scoring models may
be a useful direction for future research; see Mireshghallah
et al. (2023) for reference.
5.3. Other factors impacting performance of
DetectGPT
In this section, we explore how factors such as the size of the
mask-filling model, the number of perturbations used to estimate the expectation in Equation 1, or the data distribution
of the text to be detected impact detection quality.
Source and mask-filling model scale. Here we study the
impact of the size of the source model and mask-filling
model on DetectGPT’s performance; the results are shown
in Figure 7. In particular, the increased discrimination power
of DetectGPT for larger mask-filling models supports the
interpretation that DetectGPT is estimating the curvature | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 40 | in Figure 7. In particular, the increased discrimination power
of DetectGPT for larger mask-filling models supports the
interpretation that DetectGPT is estimating the curvature
of the log probability in a latent semantic space, rather
than in raw token embedding space. Larger T5 models
better represent this latent space, where random directions
correspond to meaningful changes in the text.
Number of perturbations for DetectGPT. We evaluate
the performance of DetectGPT as a function of the number
of perturbations used to estimate the expectation in Equation 1 on three datasets. The results are presented in Figure 8.
Detection accuracy continues to improve until 100 perturbations, where it converges. Evaluations use 100 examples
from each dataset.
Data distributional properties. We study more closely
1 10 100 10000.60.70.80.91.0Detection AUROC
GPT-2
XSum
SQuAD
WritingPrompts
1 10 100 1000
GPT-J
Number of perturbationsFigure 8. Impact of varying the number of perturbations (samples
of mask and mask-fill) used by DetectGPT on AUROC for GPT-2
(left) and GPT-J ( right ) to estimate the perturbation discrepancy | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 41 | (left) and GPT-J ( right ) to estimate the perturbation discrepancy
on detection. Averaging up to 100 perturbations greatly increases
DetectGPT’s reliability. Perturbations sampled from T5-large.
the impact of the data distribution on DetectGPT, particularly how the domain impacts the threshold separating the
perturbation discrepancy distributions of model-generated
and human texts as well as the impact of passage length on
detection. Figure 9 shows the perturbation discrepancy distributions for model-generated and human texts across four
data distributions, using GPT-Neo-2.7B to generate samples. A threshold of slightly below 0.1 separates human and
model texts across data distributions, which is important for
practical scenarios in which a passage may be analyzed without knowing its domain a priori. Finally, Figure 10 shows an
analysis of DetectGPT’s performance as a function of passage length. We bin the paired human- and model-generated
sequences by their average length into three bins of equal
size (bottom/middle/top third), and plot the AUROC within
each bin. The relationship between detection performance
and passage length generally depends on the dataset and
model (or tokenizer). For very long sequences, DetectGPT
may see reduced performance because our implementation | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 42 | and passage length generally depends on the dataset and
model (or tokenizer). For very long sequences, DetectGPT
may see reduced performance because our implementation
of DetectGPT applies all T5 mask-filling perturbations at
once, and T5 may fail to track many mask tokens at once.
By applying perturbations in multiple sequential rounds of
smaller numbers of masks, this effect may be mitigated.
6. Discussion
As large language models continue to improve, they will
become increasingly attractive tools for replacing human
writers in a variety of contexts, such as education, journalism, and art. While legitimate uses of language model
technologies exist in all of these settings, teachers, readers,
and consumers are likely to demand tools for verifying the
human origin of certain content with high educational, societal, or artistic significance, particularly when factuality
(and not just fluency) is crucial.
In light of these elevated stakes and the regular emergence of
new large language models, we study the zero-shot machinegenerated text detection problem, in which we use only the
raw log probabilities computed by a generative model to
8
Zero-Shot Machine-Generated Text Detection using Probability Curvature
0.0 0.1 0.20204060XSum
0.1 | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 43 | 8
Zero-Shot Machine-Generated Text Detection using Probability Curvature
0.0 0.1 0.20204060XSum
0.1
0.0 0.1 0.2WritingPrompts
0.0 0.1 0.20204060SQuAD
Human
Model
0.1
0.0 0.1 0.2 0.3PubMed
0.0 0.2 0.4 0.6 0.8 1.0
Log Probability Change (Perturbation Discrepancy)0.00.20.40.60.81.0Frequency
Figure 9. Perturbation discrepancy distributions for GPT-Neo
(2.7B) and humans across domains. A threshold of 0.1 generally separates model- and human-generated text well, which is
important for practical scenarios where the domain is unknown.
determine if a candidate passage was sampled from it. We
identify a property of the log probability function computed
by a wide variety of large language models, showing that a
tractable approximation to the trace of the Hessian of the
model’s log probability function provides a useful signal
for detecting model samples. Our experiments find that
this signal is more discriminative than existing zero-shot
detection methods and is competitive with bespoke detection | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 44 | for detecting model samples. Our experiments find that
this signal is more discriminative than existing zero-shot
detection methods and is competitive with bespoke detection
models trained with millions of model samples.
DetectGPT and Watermarking. One interpretation of
the perturbation function is producing semantically similar rephrasings of the original passage . If these rephrasings
are systematically lower-probability than the original passage, the model is exposing its bias toward the specific (and
roughly arbitrary, by human standards) phrasing used. In
other words, LLMs that do not perfectly imitate human
writing essentially watermark themselves implicitly. Under
this interpretation, efforts to manually add watermarking biases to model outputs (Aaronson, 2022; Kirchenbauer et al.,
2023) may further improve the effectiveness of methods
such as DetectGPT, even as LLMs continue to improve.
Limitations. One limitation of probability-based methods
for zero-shot machine-generated text detection (like DetectGPT) is the white-box assumption that we can evaluate log
probabilities of the model(s) in question. For models behind APIs that do provide probabilities (such as GPT-3),
evaluating probabilities nonetheless costs money. Another | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 45 | probabilities of the model(s) in question. For models behind APIs that do provide probabilities (such as GPT-3),
evaluating probabilities nonetheless costs money. Another
assumption of DetectGPT is access to a reasonable perturbation function. While in this work, we use off-the-shelf
mask-filling models such as T5 and mT5 (for non-English
languages), some domains may see reduced performance
if existing mask-filling models do not well represent the
space of meaningful rephrases, reducing the quality of the
curvature estimate. While DetectGPT provides the best
available detection performance for PubMedQA, its drop
Average length0.9850.9900.995AUROC
gpt-2
Average length0.960.970.980.99AUROC
opt-2.7
XSum
SQuAD
WritingPrompts
130 140 150 160 170
Average length0.8750.9000.9250.9500.975AUROC
EleutherAI/gpt-j-6b
130 140 150 160 170
Average length0.70.80.9AUROC
EleutherAI/gpt-neox-20bFigure 10. DetectGPT AUROC vs passage length. The relationship between detection performance and passage length generally | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 46 | EleutherAI/gpt-neox-20bFigure 10. DetectGPT AUROC vs passage length. The relationship between detection performance and passage length generally
depends on the dataset and model (or tokenizer). Decreases in
detection quality with increasing length may be due to T5 failing
to track many (20+) masks to fill at once; this problem may be
mitigated by applying mask-fills in a sequence of smaller batches.
in performance compared to other datasets may be a result
of lower quality perturbations. Finally, DetectGPT is more
compute-intensive than other methods for detection, as it
requires sampling and scoring the set of perturbations for
each candidate passage, rather than just the candidate passage; a better tuned perturbation function or more efficient
curvature approximation may help mitigate these costs.
Future Work. While the methods in this work make no
assumptions about the models generating the samples, future work may explore how watermarking algorithms can be
used in conjunction with detection algorithms like DetectGPT to further improve detection robustness as language
models continually improve their reproductions of human
text. Separately, the results in Section 5.2 suggest that extending DetectGPT to use ensembles of models for scoring,
rather than a single model, may improve detection in the | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 47 | rather than a single model, may improve detection in the
black box setting. Another topic that remains unexplored
is the relationship between prompting and detection; that
is, can a clever prompt successfully prevent a model’s generations from being detected by existing methods? Finally,
future work may explore whether the local log probability curvature property we identify is present for generative
models in other domains, such as audio, video, or images.
We hope that the present work serves as inspiration to future work developing effective, general-purpose methods
for mitigating potential harms of machine-generated media.
Acknowledgements
EM gratefully acknowledges funding from a KnightHennessy Graduate Fellowship. CF and CM are CIFAR
Fellows. The Stanford Center for Research on Foundation
Models (CRFM) provided part of the compute resources
9
Zero-Shot Machine-Generated Text Detection using Probability Curvature
used for the experiments in this work.
References
Aaronson, S. My Projects at OpenAI, Nov 2022. URL
https://scottaaronson.blog/?p=6823 .
Bakhtin, A., Gross, S., Ott, M., Deng, Y ., Ranzato, M.,
and Szlam, A. Real or fake? Learning to discriminate
machine from human generated text. arXiv , 2019. URL | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 48 | and Szlam, A. Real or fake? Learning to discriminate
machine from human generated text. arXiv , 2019. URL
http://arxiv.org/abs/1906.03351 .
Black, S., Gao, L., Wang, P., Leahy, C., and Biderman, S.
GPT-Neo: Large Scale Autoregressive Language Modeling with Mesh-Tensorflow, March 2021. URL https:
//doi.org/10.5281/zenodo.5297715 .
Black, S., Biderman, S., Hallahan, E., Anthony, Q., Gao,
L., Golding, L., He, H., Leahy, C., McDonell, K., Phang,
J., Pieler, M., Prashanth, U. S., Purohit, S., Reynolds, L.,
Tow, J., Wang, B., and Weinbach, S. GPT-NeoX-20B:
An open-source autoregressive language model. In Proceedings of the ACL Workshop on Challenges & Perspectives in Creating Large Language Models , 2022. URL
https://arxiv.org/abs/2204.06745 .
Bojar, O. r., Chatterjee, R., Federmann, C., Graham, Y ., | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 49 | Bojar, O. r., Chatterjee, R., Federmann, C., Graham, Y .,
Haddow, B., Huck, M., Jimeno Yepes, A., Koehn, P.,
Logacheva, V ., Monz, C., Negri, M., Neveol, A., Neves,
M., Popel, M., Post, M., Rubino, R., Scarton, C., Specia,
L., Turchi, M., Verspoor, K., and Zampieri, M. Findings
of the 2016 conference on machine translation. In Proceedings of the First Conference on Machine Translation ,
pp. 131–198, Berlin, Germany, August 2016. Association for Computational Linguistics. URL http://www.
aclweb.org/anthology/W/W16/W16-2301 .
Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan,
J. D., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry,
G., Askell, A., Agarwal, S., Herbert-V oss, A., Krueger,
G., Henighan, T., Child, R., Ramesh, A., Ziegler, D., | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 50 | G., Henighan, T., Child, R., Ramesh, A., Ziegler, D.,
Wu, J., Winter, C., Hesse, C., Chen, M., Sigler, E.,
Litwin, M., Gray, S., Chess, B., Clark, J., Berner, C.,
McCandlish, S., Radford, A., Sutskever, I., and Amodei,
D. Language models are few-shot learners. In Larochelle,
H., Ranzato, M., Hadsell, R., Balcan, M., and Lin,
H. (eds.), Advances in Neural Information Processing
Systems , volume 33, pp. 1877–1901. Curran Associates, Inc., 2020. URL https://proceedings.
neurips.cc/paper/2020/file/
1457c0d6bfcb4967418bfb8ac142f64a-Paper.
pdf.
Chowdhery, A., Narang, S., Devlin, J., Bosma, M., Mishra,
G., Roberts, A., Barham, P., Chung, H. W., Sutton, | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 51 | G., Roberts, A., Barham, P., Chung, H. W., Sutton,
C., Gehrmann, S., Schuh, P., Shi, K., Tsvyashchenko,S., Maynez, J., Rao, A., Barnes, P., Tay, Y ., Shazeer,
N., Prabhakaran, V ., Reif, E., Du, N., Hutchinson, B.,
Pope, R., Bradbury, J., Austin, J., Isard, M., Gur-Ari,
G., Yin, P., Duke, T., Levskaya, A., Ghemawat, S., Dev,
S., Michalewski, H., Garcia, X., Misra, V ., Robinson,
K., Fedus, L., Zhou, D., Ippolito, D., Luan, D., Lim,
H., Zoph, B., Spiridonov, A., Sepassi, R., Dohan, D.,
Agrawal, S., Omernick, M., Dai, A. M., Pillai, T. S.,
Pellat, M., Lewkowycz, A., Moreira, E., Child, R., Polozov, O., Lee, K., Zhou, Z., Wang, X., Saeta, B., Diaz, | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 52 | M., Firat, O., Catasta, M., Wei, J., Meier-Hellstern, K.,
Eck, D., Dean, J., Petrov, S., and Fiedel, N. PaLM:
Scaling language modeling with pathways, 2022. URL
https://arxiv.org/abs/2204.02311 .
Christian, J. CNET secretly used AI on articles that
didn’t disclose that fact, staff say. https://web.
archive.org/web/20230124063916/https:
//futurism.com/cnet-ai-articles-label ,
2023. Accessed: 2023-01-25.
Christiano, P. F., Leike, J., Brown, T., Martic, M., Legg,
S., and Amodei, D. Deep reinforcement learning from
human preferences. In Guyon, I., Luxburg, U. V .,
Bengio, S., Wallach, H., Fergus, R., Vishwanathan, S.,
and Garnett, R. (eds.), Advances in Neural Information
Processing Systems , volume 30. Curran Associates, Inc.,
2017. URL https://proceedings.neurips.
cc/paper_files/paper/2017/file/ | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 53 | Processing Systems , volume 30. Curran Associates, Inc.,
2017. URL https://proceedings.neurips.
cc/paper_files/paper/2017/file/
d5e2c0adad503c91f91df240d0cd4e49-Paper.
pdf.
Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. BERT:
Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for
Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) , pp. 4171–4186,
Minneapolis, Minnesota, June 2019. Association for
Computational Linguistics. doi: 10.18653/v1/N19-1423.
URL https://aclanthology.org/N19-1423 .
Dolhansky, B., Bitton, J., Pflaum, B., Lu, J., Howes, R.,
Wang, M., and Ferrer, C. C. The deepfake detection challenge dataset, 2020. URL https://ai.facebook.
com/datasets/dfdc/ .
Fagni, T., Falchi, F., Gambini, M., Martella, A., and | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 54 | com/datasets/dfdc/ .
Fagni, T., Falchi, F., Gambini, M., Martella, A., and
Tesconi, M. Tweepfake: About detecting deepfake tweets.
PLOS ONE , 16(5):1–16, 05 2021. doi: 10.1371/journal.
pone.0251415. URL https://doi.org/10.1371/
journal.pone.0251415 .
Fan, A., Lewis, M., and Dauphin, Y . Hierarchical neural story generation. In Proceedings of the 56th Annual Meeting of the Association for Computational Lin10
Zero-Shot Machine-Generated Text Detection using Probability Curvature
guistics (Volume 1: Long Papers) , pp. 889–898, Melbourne, Australia, July 2018. Association for Computational Linguistics. doi: 10.18653/v1/P18-1082. URL
https://aclanthology.org/P18-1082 .
Gehrmann, S., Strobelt, H., and Rush, A. GLTR: Statistical
detection and visualization of generated text. In Proceedings of the 57th Annual Meeting of the Association for
Computational Linguistics: System Demonstrations , pp. | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 55 | detection and visualization of generated text. In Proceedings of the 57th Annual Meeting of the Association for
Computational Linguistics: System Demonstrations , pp.
111–116, Florence, Italy, July 2019. Association for Computational Linguistics. doi: 10.18653/v1/P19-3019. URL
https://aclanthology.org/P19-3019 .
Guarnera, L., Giudice, O., and Battiato, S. Deepfake detection by analyzing convolutional traces. In Proceedings
of the IEEE/CVF Conference on Computer Vision and
Pattern Recognition (CVPR) Workshops , June 2020.
G¨uera, D. and Delp, E. J. Deepfake video detection
using recurrent neural networks. In 2018 15th IEEE
International Conference on Advanced Video and Signal Based Surveillance (AVSS) , pp. 1–6, 2018. doi:
10.1109/A VSS.2018.8639163.
Hutchinson, M. A stochastic estimator of the trace of
the influence matrix for laplacian smoothing splines.
Communications in Statistics - Simulation and Computation , 19(2):433–450, 1990. doi: 10.1080/
03610919008812866. URL https://doi.org/10. | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 56 | 03610919008812866. URL https://doi.org/10.
1080/03610919008812866 .
Ippolito, D., Duckworth, D., Callison-Burch, C., and Eck,
D. Automatic detection of generated text is easiest when
humans are fooled. In Proceedings of the 58th Annual
Meeting of the Association for Computational Linguistics, pp. 1808–1822, Online, July 2020. Association
for Computational Linguistics. doi: 10.18653/v1/2020.
acl-main.164. URL https://www.aclweb.org/
anthology/2020.acl-main.164 .
Jawahar, G., Abdul-Mageed, M., and Lakshmanan, L. V . S.
Automatic detection of machine generated text: A critical
survey. In International Conference on Computational
Linguistics , 2020.
Jin, Q., Dhingra, B., Liu, Z., Cohen, W., and Lu, X. PubMedQA: A dataset for biomedical research question answering. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and
the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) , pp. 2567–2577,
Hong Kong, China, November 2019. Association for | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 57 | the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) , pp. 2567–2577,
Hong Kong, China, November 2019. Association for
Computational Linguistics. doi: 10.18653/v1/D19-1259.
URL https://aclanthology.org/D19-1259 .
Kirchenbauer, J., Geiping, J., Wen, Y ., Katz, J., Miers, I.,
and Goldstein, T. A watermark for large language mod-els, 2023. URL https://arxiv.org/abs/2301.
10226 .
Krishna, K., Song, Y ., Karpinska, M., Wieting, J., and
Iyyer, M. Paraphrasing evades detectors of ai-generated
text, but retrieval is an effective defense. arXiv preprint
arXiv:2303.13408 , 2023.
Liang, W., Yuksekgonul, M., Mao, Y ., Wu, E., and Zou,
J. Gpt detectors are biased against non-native english
writers. arXiv preprint arXiv:2304.02819 , 2023. | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 58 | J. Gpt detectors are biased against non-native english
writers. arXiv preprint arXiv:2304.02819 , 2023.
Lin, S., Hilton, J., and Evans, O. TruthfulQA: Measuring
how models mimic human falsehoods. In Proceedings of
the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pp. 3214–
3252, Dublin, Ireland, May 2022. Association for Computational Linguistics. doi: 10.18653/v1/2022.acl-long.
229. URL https://aclanthology.org/2022.
acl-long.229 .
Liu, Y ., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D.,
Levy, O., Lewis, M., Zettlemoyer, L., and Stoyanov, V .
Roberta: A robustly optimized bert pretraining approach.
arXiv preprint arXiv:1907.11692 , 2019.
Mireshghallah, F., Mattern, J., Gao, S., Shokri, R., and BergKirkpatrick, T. Smaller language models are better blackbox machine-generated text detectors. arXiv preprint | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 59 | arXiv:2305.09859 , 2023.
Narayan, S., Cohen, S. B., and Lapata, M. Don’t give me
the details, just the summary! Topic-aware convolutional
neural networks for extreme summarization. In Proceedings of the 2018 Conference on Empirical Methods in
Natural Language Processing , Brussels, Belgium, 2018.
OpenAI. Chatgpt: Optimizing language models for
dialogue. http://web.archive.org/web/
20230109000707/https://openai.com/
blog/chatgpt/ , 2022. Accessed: 2023-01-10.
Radford, A., Wu, J., Child, R., Luan, D., Amodei,
D., and Sutskever, I. Language models are
unsupervised multitask learners, 2019. URL
https://d4mucfpksywv.cloudfront.net/
better-language-models/language_
models_are_unsupervised_multitask_
learners.pdf .
Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 60 | learners.pdf .
Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang,
S., Matena, M., Zhou, Y ., Li, W., and Liu, P. J. Exploring the limits of transfer learning with a unified
text-to-text transformer. Journal of Machine Learning
Research , 21(140):1–67, 2020. URL http://jmlr.
org/papers/v21/20-074.html .
11
Zero-Shot Machine-Generated Text Detection using Probability Curvature
Rajpurkar, P., Zhang, J., Lopyrev, K., and Liang, P. SQuAD:
100,000+ questions for machine comprehension of text.
InProceedings of the 2016 Conference on Empirical
Methods in Natural Language Processing , pp. 2383–
2392, Austin, Texas, November 2016. Association for
Computational Linguistics. doi: 10.18653/v1/D16-1264.
URL https://aclanthology.org/D16-1264 .
Roose, K. and Newton, C. ChatGPT transforms a
classroom and is ‘M3GAN’ real? Hard Fork, a
New York Times Podcast, 2022. URL https:// | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 61 | classroom and is ‘M3GAN’ real? Hard Fork, a
New York Times Podcast, 2022. URL https://
www.nytimes.com/2023/01/13/podcasts/
hard-fork-chatgpt-teachers-gen-z-cameras-m3gan.
html .
Sadasivan, V . S., Kumar, A., Balasubramanian, S., Wang,
W., and Feizi, S. Can ai-generated text be reliably detected? arXiv preprint arXiv:2303.11156 , 2023.
Solaiman, I., Brundage, M., Clark, J., Askell, A., HerbertV oss, A., Wu, J., Radford, A., and Wang, J. Release
strategies and the social impacts of language models,
2019. URL https://arxiv.org/ftp/arxiv/
papers/1908/1908.09203.pdf .
Uchendu, A., Le, T., Shu, K., and Lee, D. Authorship attribution for neural text generation. In Proceedings of the 2020
Conference on Empirical Methods in Natural Language
Processing (EMNLP) , pp. 8384–8395, Online, November
2020. Association for Computational Linguistics. doi: | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 62 | Processing (EMNLP) , pp. 8384–8395, Online, November
2020. Association for Computational Linguistics. doi:
10.18653/v1/2020.emnlp-main.673. URL https://
aclanthology.org/2020.emnlp-main.673 .
Wang, B. and Komatsuzaki, A. GPT-J-6B: A
6 Billion Parameter Autoregressive Language
Model. https://github.com/kingoflolz/
mesh-transformer-jax , May 2021.
Zellers, R., Holtzman, A., Rashkin, H., Bisk, Y ., Farhadi,
A., Roesner, F., and Choi, Y . Defending against neural
fake news. In Neural Information Processing Systems ,
2019.
Zhang, S., Roller, S., Goyal, N., Artetxe, M., Chen, M.,
Chen, S., Dewan, C., Diab, M., Li, X., Lin, X. V ., Mihaylov, T., Ott, M., Shleifer, S., Shuster, K., Simig, D.,
Koura, P. S., Sridhar, A., Wang, T., and Zettlemoyer, | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 63 | Koura, P. S., Sridhar, A., Wang, T., and Zettlemoyer,
L. Opt: Open pre-trained transformer language models, 2022. URL https://arxiv.org/abs/2205.
01068 .
Zhao, H., Zhou, W., Chen, D., Wei, T., Zhang, W., and Yu, N.
Multi-attentional deepfake detection. In Proceedings of
the IEEE/CVF conference on computer vision and pattern
recognition , pp. 2185–2194, 2021.Zi, B., Chang, M., Chen, J., Ma, X., and Jiang, Y .-G. Wilddeepfake: A challenging real-world dataset for deepfake
detection. In Proceedings of the 28th ACM international
conference on multimedia , pp. 2382–2390, 2020.
Ziegler, D. M., Stiennon, N., Wu, J., Brown, T. B., Radford,
A., Amodei, D., Christiano, P., and Irving, G. Fine-tuning
language models from human preferences, 2020.
12
Zero-Shot Machine-Generated Text Detection using Probability Curvature
A. Complete Results for Top- pand Top- kDecoding | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 64 | language models from human preferences, 2020.
12
Zero-Shot Machine-Generated Text Detection using Probability Curvature
A. Complete Results for Top- pand Top- kDecoding
Tables 4 and 5 contain the complete results for XSum, SQuAD, and WritingPrompts for the five models considered in
Table 1. On average, both top- pand top- ksampling seem to make the detection task easier. This result is perhaps intuitive,
as both sampling methods strictly increase the average log likelihood of model generations under the model (as they truncate
low-probability tokens, albeit with different heuristics). Therefore methods based on probability or rank of tokens should
become more discriminative.
XSum SQuAD WritingPrompts
Method GPT-2 OPT-2.7 Neo-2.7 GPT-J NeoX Avg. GPT-2 OPT-2.7 Neo-2.7 GPT-J NeoX Avg. GPT-2 OPT-2.7 Neo-2.7 GPT-J NeoX Avg.
logp(x) 0.93 0.93 0.94 0.91 0.87 0.92 0.96 0.94 0.91 0.87 0.79 0.89 0.99* 0.98* 0.98* 0.97* 0.97* 0.98 | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 65 | Rank 0.80 0.77 0.77 0.75 0.73 0.76 0.84 0.82 0.81 0.80 0.75 0.81 0.87 0.84 0.83 0.83 0.81 0.84
LogRank 0.95* 0.94* 0.96* 0.93* 0.89* 0.93* 0.98* 0.96* 0.94* 0.90 0.83 0.92* 0.99* 0.98* 0.98* 0.98 0.98 0.98
Entropy 0.55 0.46 0.53 0.54 0.58 0.53 0.53 0.50 0.55 0.56 0.57 0.54 0.32 0.37 0.28 0.32 0.32 0.32
DetectGPT 0.99 0.98 1.00 0.98 0.97 0.98 0.99 0.98 0.98 0.90 0.82* 0.94 1.00 0.99 0.99 0.97* 0.93 0.98 | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 66 | Diff 0.04 0.04 0.04 0.05 0.08 0.05 0.01 0.02 0.04 0.00 -0.01 0.02 0.01 0.01 0.01 -0.01 -0.05 0.00
Table 4. Nucleus (top- p) sampling evaluation with p= 0:96. AUROC for detecting samples from the given model on the given dataset for
DetectGPT and four previously proposed criteria. Nucleus sampling generally makes detection easier for all methods, but DetectGPT still
provides the highest average AUROC. For WritingPrompts, however, the LogRank baseline performs as well as DetectGPT.
XSum SQuAD WritingPrompts
Method GPT-2 OPT-2.7 Neo-2.7 GPT-J NeoX Avg. GPT-2 OPT-2.7 Neo-2.7 GPT-J NeoX Avg. GPT-2 OPT-2.7 Neo-2.7 GPT-J NeoX Avg.
logp(x) 0.89 0.89 0.89 0.84 0.81 0.87 0.93 0.90 0.88 0.82 0.74 0.85 0.97 0.95 0.97 0.96 0.95* 0.96 | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 67 | Rank 0.79 0.77 0.77 0.75 0.73 0.76 0.84 0.82 0.80 0.80 0.75 0.80 0.87 0.84 0.83 0.82 0.81 0.83
LogRank 0.92* 0.91* 0.93* 0.89* 0.85* 0.90* 0.96* 0.94* 0.92* 0.87* 0.79* 0.90* 0.98* 0.97* 0.98* 0.97 0.96 0.97
Entropy 0.58 0.49 0.55 0.56 0.59 0.55 0.55 0.52 0.56 0.56 0.58 0.56 0.35 0.41 0.30 0.34 0.37 0.35
DetectGPT 0.99 0.97 0.99 0.96 0.96 0.98 0.99 0.98 0.98 0.89 0.80 0.93 0.99 0.98 0.99 0.97 0.93 0.97
Diff 0.07 0.06 0.06 0.07 0.11 0.08 0.03 0.04 0.06 0.02 0.01 0.03 0.01 0.01 0.01 0.00 -0.03 0.00 | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2301.11305 | 68 | Table 5. Top-ksampling evaluation with k= 40 . DetectGPT generally provides the most accurate performance (highest AUROC),
although the gap is narrowed comparing to direct sampling, presumably because top- kgenerations are more generic.
13 | 2301.11305 | DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature | The increasing fluency and widespread usage of large language models (LLMs)
highlight the desirability of corresponding tools aiding detection of
LLM-generated text. In this paper, we identify a property of the structure of
an LLM's probability function that is useful for such detection. Specifically,
we demonstrate that text sampled from an LLM tends to occupy negative curvature
regions of the model's log probability function. Leveraging this observation,
we then define a new curvature-based criterion for judging if a passage is
generated from a given LLM. This approach, which we call DetectGPT, does not
require training a separate classifier, collecting a dataset of real or
generated passages, or explicitly watermarking generated text. It uses only log
probabilities computed by the model of interest and random perturbations of the
passage from another generic pre-trained language model (e.g., T5). We find
DetectGPT is more discriminative than existing zero-shot methods for model
sample detection, notably improving detection of fake news articles generated
by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline
to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code,
data, and other project information. | http://arxiv.org/pdf/2301.11305 | [
"Eric Mitchell",
"Yoonho Lee",
"Alexander Khazatsky",
"Christopher D. Manning",
"Chelsea Finn"
] | [
"cs.CL",
"cs.AI"
] | ICML 2023 | null | cs.CL | 20230126 | 20230723 | [
{
"id": "2305.09859"
},
{
"id": "2303.11156"
},
{
"id": "2303.13408"
},
{
"id": "2301.11305"
},
{
"id": "1907.11692"
},
{
"id": "2304.02819"
}
] |
2209.07686 | 0 | October 14, 2022
TEXT AND PATTERNS : FOREFFECTIVE CHAIN OF THOUGHT
ITTAKES TWO TO TANGO
Aman Madaanand Amir Yazdanbakhsh
Carnegie Mellon University Google Research, Brain Team
[email protected], [email protected]
(Equal Contribution)
ABSTRACT
In the past decade, we witnessed dramatic gains in natural language processing and an
unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (C OT) prompting.
Specifically, C OT pushes the performance of large language models in a few-shot setup by
augmenting the prompts with intermediate steps. Despite impressive results across various
tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of C OT-based few-shot prompting
mechanisms in large language models. We first systematically identify and define the
key components of a prompt: symbols ,patterns , and text. Then, we devise and conduct
an exhaustive set of deliberated experiments across four different tasks, by querying the
model with counterfactual prompts where only one of these components is altered. Our | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 1 | an exhaustive set of deliberated experiments across four different tasks, by querying the
model with counterfactual prompts where only one of these components is altered. Our
experiments across three models—PaLM, GPT-3, and C ODEX —reveal several surprising
findings and brings into question the conventional wisdom around few-shot prompting.
First, the presence of factual patterns in a prompt is practically immaterial to the success
of C OT. Second, our results conclude that the primary role of intermediate steps may not
be to facilitate learning “how” to solve a task. The intermediate steps are rather a beacon
for the model to realize “what” symbols to replicate in the output to form a factual answer.
As such, the patterns are merely a channel to “trick” the model into forming sentences
that resemble correct answers. This pathway is facilitated by text, which imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis
reveals that a symbiotic relationship between text and patterns explains the success of fewshot prompting: text helps extract commonsense from the question to help patterns, and
patterns enforce task understanding and direct text generation. Such systematic understanding of C OT enables us to devise a concise chain of thought, dubbed as CC OT, where
text and patterns are pruned by over 20 %, only retaining their key roles. We achieve this | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 2 | text and patterns are pruned by over 20 %, only retaining their key roles. We achieve this
reduction in the number of tokens while delivering on par or slightly higher solve task rate.
Work done when Aman Madaan was a student researcher at Google Research, Brain Team.
1. I NTRODUCTION
The ability to learn a previously unseen task by observing a few examples is one of the cornerstones of
human intelligence (Lake et al., 2017). This is in stark contrast with modern deep learning methods, which
typically rely on a substantial labeled corpus of data. Recently, large language models (LLMs) (Chowdhery
et al., 2022; Brown et al., 2020; Chen et al., 2021a) have demonstrated remarkable performance in employing
aprompt to perform a task, with no additional finetuning, commonly known as few-shot learning. Few-shot
1arXiv:2209.07686v2 [cs.CL] 13 Oct 2022
learning has shown promising applications for a wide range of tasks (Gehrmann et al., 2021; Wei et al.,
2021; Sanh et al., 2021; Thoppilan et al., 2022; Liu et al., 2021a; Reif et al., 2021; Wang et al., 2020; Chen | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 3 | 2021; Sanh et al., 2021; Thoppilan et al., 2022; Liu et al., 2021a; Reif et al., 2021; Wang et al., 2020; Chen
et al., 2021b; Lewkowycz et al., 2022; Wu et al., 2022). While beneficial, this setting requires meticulous
design of prompts (Le Scao & Rush, 2021; Liu et al., 2021c; Mishra et al., 2021).
Ling et al. (2017) pioneered the idea of using natural language rationales as the intermediate steps in prompts
to help model performance for mathematical reasoning. Recently, Wei et al. (2022) proposed chain of
thought (C OT) prompting, showing that the few-shot setting in LLMs similarly benefits from intermediate
natural language rationale across a range of complex reasoning tasks (Ling et al., 2017; Cobbe et al., 2021;
Patel et al., 2021; BIG-bench Collaboration, 2022). Despite its wide-range usage, the rationale behind the
success of C OT remains unclear. Recent work draws (Ling et al., 2017; Wei et al., 2022) parallels to human
thinking. Humans often think about a problem before deducing a solution. Akin to this process, it is argued
that models should also be able to employ a similar mechanism. While intuitive, such restrictive abstract | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 4 | that models should also be able to employ a similar mechanism. While intuitive, such restrictive abstract
explanations fall short in explaining why,when , and how these mechanisms operate. Ultimately, LLMs
are trained to estimate the next token distribution for a given context. Therefore, there is presumably a
systematic rationale behind their successes and failures. In this work, we undertake initial steps towards
understanding the mechanism behind C OT.
Contributions and findings. We construct a series of tailored counterfactual prompts (Goyal et al., 2019),
deliberately sketched as controlled studies. First, we identify key components of an example in few-shot
prompting as follows: Symbols ,Patterns , and Text. Next, we perform counterfactual prompting —keeping
all but one component fixed (e.g., replacing symbols (numbers) with Greek alphabets). Finally, we elicit
meaningful findings via conducting a systematic and qualitative analysis of the performance divergence
between different prompt queries. Our experiments on four diverse reasoning tasks and across three large
language models—PaLM, GPT-3, and C ODEX , reveal several surprising findings:
1We find that the exact type of symbols in the prompt virtually does not affect the model performance. In | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 5 | 1We find that the exact type of symbols in the prompt virtually does not affect the model performance. In
addition, our results and analysis demonstrate counterintuitive phenomena. For example, we identify that the
correctness of symbols and patterns is immaterial to the task solve rate. 2We learn that patterns contribute
chiefly as a venue to reinforce task understanding (Ouyang et al., 2022) and prompt the model to attain
correct outputs. 3Most importantly, we find that text and patterns form a symbiotic relationship that plays
a vital role in the success of C OT. Text helps generate useful patterns (e.g., by extracting commonsense
knowledge), and patterns help reinforce task understanding, enabling the language model to generate text
that helps solve the task. Overall, we argue that one of the primary reasons behind the success of C OT is this
interplay between text and patterns—C OT helps a language model in imitating the prompt and generating
the right tokens for the task—and is conceivably less related to their reasoning abilities. Finally, as indicated
by applications such as PaLM-S AYCAN (Ahn et al., 2022), we posit that techniques like C OT will play a key
role in enabling the success of LLMs on diverse use cases. Thus, designing efficient prompts informed by a | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 6 | role in enabling the success of LLMs on diverse use cases. Thus, designing efficient prompts informed by a
set of key design principles is an important challenge. To this end, we distill our findings to create concise
prompting, dubbed CC OT. CC OT prunes the prompt (20 %Ó) to only retain indispensable tokens without
negative repercussions on the task solve rate.
2. C OUNTERFACTUAL EXPLANATION FOR CHAIN OF THOUGHT
The primary objective of our study is to understand C OT through counterfactual prompting and empirically
establish the underpinnings of the reasoning ability of LLMs in the presence of C OT. Each counterfactual
promptCfppqalters only one particular aspect of the in-context examples xxktkykyin ap. For example,
consider a sample thought for the math world problems in the GSM-8 Kdataset (See Table 1). A symbolic
counterfactual prompt, Csymbolicppq, may simply replace all the numbers in the thoughts with symbols (e.g.,
X1). Such analysis enables us to ask: “ what would theperformance ofthemodel have been, ifallthe
numbers intheprompt were replaced with symbols? ”. Analyzing the performance disparity of a LLM on
2 | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 7 | numbers intheprompt were replaced with symbols? ”. Analyzing the performance disparity of a LLM on
2
Csymbolicppqvs.pcan thus indicate the role that using actual numbers plays in the success or failure of a
task1. The ability to successfully complete prompts pfor complex reasoning tasks is typically present for
LLM at the scale of PaLM, GPT-3, and C ODEX . Nonetheless, we do not make any assumptions about the
underlying model architecture. In summary, our study on GSM-8 Kreveals that for solving math problems,
neither the presence of numbers, nor the credibility of the thoughts is paramount to the success of C OT.
Similarly, altering the style and wording of the texts in the thoughts has a modest impact on the model
performance. Nevertheless, eradicating either of these components nullifies the efficacy of C OT. Finally,
a per-layer analysis of the model reveals that if the model performs similarly for a pair of counterfactual
promptsCfppqvs.p, then the attention patterns are comparable as well. We defer the detailed background
to Appendix A.
Limitations of counterfactual prompting. Relying on counterfactual examples could be misleading and | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 8 | to Appendix A.
Limitations of counterfactual prompting. Relying on counterfactual examples could be misleading and
precarious (Laugel et al., 2019; Slack et al., 2021). Nonetheless, counterfactual explanation presents a
channel to gain insights into the workings of the model. This approach potentially yields more favorable
explanations for state-of-the-art LLMs. Notably, unlike fine-tuned methods, one can readily identify and
collect a set of prompts that are critical for the model to generate particular outputs. In particular, fewshot prompting augments the model with an additional dimension to calibrate the accuracy to a discernible
degree. Thus, we deduce that the counterfactual examples that exhibit consistent andsystematic performance
divergence are more prone to reflect credible interpretations of the model. In this work, we neither rely on
the results that do not exhibit such characteristics, nor reject prompts that pose contradictory observations.
We discuss additional limitations in Appendix A.1.
3. E XPERIMENTAL SETUP
Large language models. To facilitate conducting an exhaustive number of experiments, we center the main
analysis of this paper around PaLM-62 B2. For reproducibility, we also conduct the experiments on publicly
available models such as GPT-3 and C ODEX . We present results from PaLM-62 Bin the main body and | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 9 | available models such as GPT-3 and C ODEX . We present results from PaLM-62 Bin the main body and
defer the results from other models to Appendix E. Nonetheless, our findings concur across the studied
LLMs.
Reasoning tasks. We focus on reasoning tasks for which C OT presents ample improvements
over D IRECT prompting (Wei et al., 2022), namely MATHEMATICAL (GSM-8 KCobbe et al.
(2021)), COMMONSENSE (date and sports understanding BIG-bench Collaboration (2022)), and
SYMBOLIC (SORTING )—details in Appendix B and Appendix-Table 10.
3.1. S EMANTIC COMPONENTS OF PROMPTS
This work intends to tease apart the major semantic components of a prompt that play a critical role in the
efficacy of C OT. To achieve this, we identify and systematically construe three key semantic components of
a prompt, listed as follows (See Appendix-Table 10 for the examples to which we refer in the definitions):
Symbols are sequences of tokens in the prompt, about which the model reasons to solve a task. For GSM-8 Kand | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 10 | Symbols are sequences of tokens in the prompt, about which the model reasons to solve a task. For GSM-8 Kand
SORTING , symbols are numerical quantities (e.g., 5, 4, 2, 13). Similarly for the S PORTS dataset, we
categorize players and activities as symbols. We define the symbols in the D ATEdataset as date and time
indicating expressions.
Patterns are either composition of symbols and operators or a structure of prompt that reinforces task understanding.The isolation of patterns within a prompt is evident in datasets like GSM-8 K(equations), S PORTS
1Appendix-Table 9 outlines a set of major “ what if ” questions that we study in this work.
2The results in Wei et al. (2022) reinforce that PaLM-62 Bis the smallest model with significant gains across a range of
reasoning tasks.
3
Table 1: Symbols ,Patterns , andTextacross different tasks.
MATHEMATICAL
Question: Shawn has five toys. For Christmas, he got two toys each from his mom and dad. How many toys does he have now? | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 11 | Question: Shawn has five toys. For Christmas, he got two toys each from his mom and dad. How many toys does he have now?
Thought: Shawn started with 5toys. If he got 2toys each from his mom and dad, then that is 4more toys. 55+44=99.
COMMONSENSE (SPORTS )
Question: Is the following sentence plausible? "Jamal Murray was perfect from the line."’
Thought: Jamal Murray is abasketball player. Being perfect from the line is part of basketball .
COMMONSENSE (DATE)
Question: It is4/19/1969 today. What is the date 24 hours later in MM/DD/YYYY?
Thought:xcalculationyToday is 04/19/1969 . 24 hours later is one day after today, which would be 04/20/1969 .xoutputyThe answer is
04/20/1969 .
SYMBOLIC (SORTING )
Question : 3, 1, 2, 7, 8, 5, 6, 9, 4
Thought: 11<22:::<99
(person is a sport, activity is a sport ), and S ORTING (1 less than 2) . However, for the D ATE dataset, | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 12 | (person is a sport, activity is a sport ), and S ORTING (1 less than 2) . However, for the D ATE dataset,
the pattern is semantically implicit but consistent. Each thought contains two parts: (a) xcalculationyin
which the information from the input is restated (e.g., “Today is 06/02/1943”) and intermediate results
are generated (e.g., “One day after 06/01/1943 is 06/02/1943”) through mathematical calculations and
(b)xoutputyin which the final requisite answer is generated using the intermediate results (e.g., “10
days before today is 05/23/1943”).
Textare tokens that are neither symbols, nor part of patterns. Specifically, text in prompts assists in either outlining the target task (e.g., is the sentence plausible ), connecting patterns to symbols (e.g.,
John is left with 4 - 2 = 2), or contextualizing symbols (4 toys). In a nutshell, text is the conceptual
glue that binds different parts of a prompt.
Relying on our analysis, we conjecture and hypothesize about the effects of each semantic components
on the outcomes of C OT prompting. We discuss detailed results, including statistical significance tests | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 13 | on the outcomes of C OT prompting. We discuss detailed results, including statistical significance tests
in Appendix E.
4. R OLE OF SYMBOLS
«H0» The exact value and type of symbols are mostly immaterial to the model performance. Replacing
symbols with abstract placeholders can do just as well at eliciting effective thoughts.
Visually inspecting the examples in Table 1, it seems intuitive to assume that symbols are important for
steering the model towards comprehending (or presumably reasoning about) a target task. We form a set of
counterfactual prompts in which the symbols are deliberately altered in distinct ways (Table 2 and AppendixTable 19 summarize the results). To test this hypothesis, we conducted two sets of experiments using counterfactual prompts: replacing the symbols with abstract values, and replacing them with out-of-distribution
symbols.
Abstract symbols vCsymb_absppqw.We first experiment with the role of symbols by creating Csymb_absppq, a
modified variant of prompt pin which some or all the symbols are replaced with an abstract placeholder (See
Table 2). The results in Table 2 illustrate that the performance has little to no impact when the symbols are
replaced with abstract placeholders. Note that for the S PORTS dataset, we also experiment with changing | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 14 | replaced with abstract placeholders. Note that for the S PORTS dataset, we also experiment with changing
sportsperson and sport activity, which mutates the baseline thoughts to vague and ungrammatical and drops
the task rate to 52.96% (Table 20).
4
Table 2: A sample modified thought for each category is depicted below. We accordingly update the
questions associated with each thought. Appendix-Table 19 provides additional results.
Question / Thought Prompt Type Solve Rate
MATHEMATICAL (DIRECT = 10.11 %, COT = 27.37 %)
Thought: Shawn started with toys. If he got toys each from his mom and dad,
then that ismore toys. +=.Csymb_absppq(Table 43) 25.70 %
Thought: Shawn started with 5.5toys. If he got 2.5toys each from his mom and dad,
then that is 5more toys. 5.5+5=10.5.Csymb_oodppq(Table 48) 28.20 %
COMMONSENSE (SPORTS ) (D IRECT = 71.08 %, COT = 93.67 %) | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 15 | COMMONSENSE (SPORTS ) (D IRECT = 71.08 %, COT = 93.67 %)
Thought: Jamal Murray is a basketball player. Being ACTIVITY is part of basketball. Csymb_absppq(Table 46) 92.11 %
Thought: Adair Foster is a basketball player. Juggling the paper cups is part of
basketball.Csymb_oodppq(Table 50) 79.72 %
COMMONSENSE (DATE) (D IRECT = 31.61 %, COT = 45.18 %)
Thought: Today is DATE . 24 hours later is one day after today, which would be DATE .Csymb_absppq(Table 42) 37.41 %
Thought: Today is 04/30/3069 . 24 hours later is one day after today, which would be
04/31/3069 .Csymb_oodppq(Table 49) 44.50 %
SYMBOLIC (SORTING ) (D IRECT = 46.0 %, COT = 60.6 %)
Thought: c< <
<<< <<< Csymb_absppq(Table 44) 61.8 % | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 16 | Thought: c< <
<<< <<< Csymb_absppq(Table 44) 61.8 %
Thought: 11 23 34 48 56 63 72 85 95 Csymb_oodppq(Table 51) 80.0 %
Out of distribution symbols vCsymb_oodppqw.To test the operational utility of symbols, we design counterfactual prompts Csymb_oodppq, in which the symbols are sampled from a distinct distribution compared to
the symbols in the questions. The operations include replacing integers in GSM-8 Kprompt with fractions,
sportsperson in S PORTS prompt with random names, and changing dates in D ATEto dates after 3000 AD.
The results (Table 2) fail to reject our hypothesis and reinforce our initial finding that the type of symbols is
primarily immaterial to the model performance. A notable exception is S PORTS , where including artificial
names and activities closes the gap between D IRECT and C OT. However, surprisingly, even with entirely
artificial names and activities in the S PORTS dataset, the model performance is marginally better than direct. | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 17 | artificial names and activities in the S PORTS dataset, the model performance is marginally better than direct.
Another interesting exception occurs in the S ORTING dataset. Compared to vanilla C OT, using larger integers (¥10) considerably improves the task solve rate (60.6 %Ñ80.0%). We postulate that in this scenario,
the modified thoughts more effectively inform the model about the underlying task of sorting numbers. These
results indicate that placeholders and abstract values can do merely as well at eliciting effective thoughts.
However, we find that completely deleting the symbols is not a viable option, as expected. Deleting all the
symbols (e.g., numbers and dates) nullifies the gains carried over by C OT. We show additional results in
Appendix-Table 15.
Nature of generated answers. We observe that the task solve rates are relatively unaffected for both
Csymb_absppqandCsymb_oodppq. In hindsight, it is not apparent whether systematic differences exist in the
generated answers. To quantify this, we compute the Cohen’s agreement score (Cohen, 1960) between
predictions generated by p(DIRECT ) and various counterfactual prompts. The results (Appendix E.2) show | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 18 | predictions generated by p(DIRECT ) and various counterfactual prompts. The results (Appendix E.2) show
that there is a moderate (>0.4) to substantial (>0.61) agreement between C OT and symbolic counterfactual
prompts. In contrast, the agreement between D IRECT and the counterfactual prompts is meager (<0.2).
These results reinforce our finding that the model may behave similarly regardless of the actual type/value
of the symbols.
Analysis of employing in-distribution symbols in thoughts. We also delve into the details of generated
answers for GSM-8 KusingpandCsymb_oodppq. As Table 2 delineates, Csymb_oodppqprompts for GSM-8 K
contain questions/thoughts with simple decimals. We investigate whether such prompts help to improve the
solve rate for questions with decimals preferentially. Surprisingly, we observe that such prompts did not
5
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9(a) GSM-8 K:pvs.Csymb_absppq
Figure 1: The average attention per token for a randomly sampled question using vanilla C OT prompt | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 19 | Figure 1: The average attention per token for a randomly sampled question using vanilla C OT prompt
p(above) and different counterfactual prompts (bottom). Near identical attention scores indicates
that few-shot models are relatively indifferent to the exact symbols, but are sensitive to patterns. In
addition, this study suggests that the model has a tendency to more profoundly attend to tokens at
the vicinity of final question (brighter bars at the right side of each bar). Please see Appendix D for
details on attention score calculation, and per-layer heatmaps. Appendix Figure 12 depicts additional
results.
Q
:
There
are
5
trees
the
grove
.
Grove
workers
will
plant
trees
in
grove
today
.
After
they
are
done
,
there
will
be
2
1
trees
.
How
many
trees
did
the
grove
workers
plant
today
?
A
: | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 20 | trees
did
the
grove
workers
plant
today
?
A
:
There
are
5
trees
originally
.
Then
there
were
2
1
trees
after
more
were
planted
.
So
there
must
have
been
2
5
=
.
The
answer
is
6
.
Q
:
If
there
are
3
cars
in
the
parking
lot
and
2
more
cars
arrive
,
how
many
cars
are
in
the
parking
lot
?
A
:
There
are
originally
3 | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 21 | parking
lot
?
A
:
There
are
originally
3
cars
.
2
more
cars
arrive
.
3
+
2
=
5
.
The
answer
is
5
.
Q
:
Leah
had
3
2
chocolates
and
her
sister
had
4
2
.
If
they
at
e
3
,
how
many
pieces
do
they
have
left
total
?
A
:
Originally
,
Leah
had
3
2
chocolates
.
Her
sister
had
4
2
.
So | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 22 | chocolates
.
Her
sister
had
4
2
.
So
total
they
had
3
2
+
4
2
=
7
4
.
After
eating
3
5
they
had
4
3
5
=
3
9
.
The
answer
is
3
9
.
Q
:
Jason
had
2
0
lollipops
.
He
gave
Denny
some
lollipops
.
Now
Jason
has
1
2
lollipops
.
How
many
lollipops
did
Jason
give
to
Denny
?
A | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 23 | many
lollipops
did
Jason
give
to
Denny
?
A
:
Jason
started
with
2
lollipops
.
Then
he
had
2
after
giving
some
to
Denny
.
So
he
gave
Denny
2
0
1
2
=
8
.
The
answer
is
8
.
Q
:
Shawn
has
five
toys
.
For
Christmas
,
he
got
two
toys
each
from
his
mom
and
dad
.
How
many
toys
does
he
have
now
?
A | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 24 | .
How
many
toys
does
he
have
now
?
A
:
Shawn
started
with
5
toys
.
If
he
got
2
toys
each
from
his
mom
and
dad
,
then
that
is
4
more
toys
.
5
+
4
=
9
.
The
answer
is
9
.
Q
:
There
were
nine
computers
in
the
server
room
.
Five
more
computers
were
installed
each
day
,
from
monday
to
thursday
.
How
many
computers | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 25 | day
,
from
monday
to
thursday
.
How
many
computers
are
now
in
the
server
room
?
A
:
There
were
originally
9
computers
.
For
each
of
4
days
,
5
more
computers
were
added
.
So
5
*
4
=
2
0
computers
were
added
.
9
+
2
0
is
2
9
.
The
answer
is
2
9
.
Q
:
Michael
had
5
8
golf
balls
.
On
tuesday
, | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 26 | Michael
had
5
8
golf
balls
.
On
tuesday
,
he
lost
2
3
golf
balls
.
On
wednesday
,
he
lost
2
more
.
How
many
golf
balls
did
he
have
at
the
end
of
wednesday
?
A
:
Michael
started
with
5
8
golf
balls
.
After
losing
2
3
on
tuesday
,
he
had
5
8
2
3
=
3
5
.
After
losing
2
more
,
he
had
3
5 | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 27 | 5
.
After
losing
2
more
,
he
had
3
5
2
=
3
3
golf
balls
.
The
answer
is
3
3
.
Q
:
Olivia
has
$
2
3
.
She
bought
five
bagels
for
$
3
each
.
How
much
money
does
she
have
left
?
A
:
Olivia
had
2
3
dollars
.
5
bagels
for
3
dollars
each
will
be
5
x
3
=
1
5
dollars | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 28 | will
be
5
x
3
=
1
5
dollars
.
So
she
has
2
3
1
5
dollars
left
.
2
3
1
5
is
8
.
The
answer
is
8
.
(a) Vanilla C OT Promptp.
Q
:
There
are
α
trees
in
the
grove
.
Grove
workers
will
plant
trees
in
the
grove
today
.
After
they
are
done
,
there
will
be
β
one
trees
.
How
many
trees
did
the
grove | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 29 | β
one
trees
.
How
many
trees
did
the
grove
workers
plant
today
?
A
:
There
are
α
trees
originally
.
Then
there
were
β
one
trees
after
some
more
were
planted
.
So
there
must
have
been
β
one
α
=
λ
.
The
answer
is
λ
.
Q
:
If
there
are
α
cars
in
the
parking
lot
and
β
more
cars
arrive
,
how
many
cars
are
the
parking | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 30 | cars
arrive
,
how
many
cars
are
the
parking
lot
?
A
:
There
are
originally
α
cars
.
β
more
cars
arrive
.
α
+
β
=
λ
.
The
answer
is
λ
.
Q
:
Leah
had
α
chocolates
and
her
sister
had
β
.
If
they
at
e
λ
,
how
many
pieces
do
they
have
left
in
total
?
A
:
Originally
,
Leah
had
α
chocolates
.
Her | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 31 | :
Originally
,
Leah
had
α
chocolates
.
Her
sister
had
β
.
So
in
total
they
had
α
+
β
=
π
.
After
eating
λ
,
they
had
π
λ
=
μ
.
The
answer
is
μ
.
Q
:
Jason
had
α
lollipops
.
He
gave
Denny
some
.
Now
Jason
has
β
lollipops
.
How
many
lollipops
did
Jason
give
to
Denny
?
A
:
Jason
started | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 32 | did
Jason
give
to
Denny
?
A
:
Jason
started
with
α
lollipops
.
Then
he
had
β
after
giving
some
to
Denny
.
So
he
gave
Denny
α
β
=
λ
.
The
answer
is
λ
.
Q
:
Shawn
has
α
toys
.
For
Christmas
,
he
got
β
toys
each
from
his
mom
and
dad
.
How
many
toys
does
he
have
now
?
A
:
Shawn
started
with
α | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 33 | he
have
now
?
A
:
Shawn
started
with
α
toys
If
he
got
β
toys
each
from
his
mom
and
dad
,
then
that
is
λ
more
toys
.
α
+
λ
=
π
.
The
answer
is
π
.
Q
:
There
were
α
computers
in
the
server
room
.
β
more
computers
were
installed
each
day
,
from
monday
to
thursday
.
How
many
computers
are
now
in
the
server
room | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 34 | .
How
many
computers
are
now
in
the
server
room
?
A
:
There
were
originally
α
computers
.
For
each
of
four
days
,
β
more
computers
were
added
.
So
β
*
four
=
λ
computers
were
added
.
α
+
λ
is
π
.
The
answer
is
π
.
Q
:
Michael
had
α
golf
balls
.
On
tuesday
,
he
lost
β
golf
balls
.
On
wednesday
,
he
lost | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 35 | lost
β
golf
balls
.
On
wednesday
,
he
lost
λ
more
.
How
many
golf
balls
did
he
have
at
the
end
of
wednesday
?
A
:
Michael
started
with
α
golf
balls
.
After
losing
β
on
tuesday
,
he
had
α
β
=
π
.
After
losing
λ
more
,
he
had
π
λ
=
μ
golf
balls
.
The
answer
is
μ
.
Q
:
Olivia
has
$
α | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 36 | answer
is
μ
.
Q
:
Olivia
has
$
α
.
She
bought
five
bagels
for
$
β
each
.
How
much
money
does
she
have
left
?
A
:
Olivia
had
α
dollars
.
5
bagels
for
β
dollars
each
will
be
5
x
β
=
λ
dollars
.
So
she
has
α
λ
dollars
left
.
α
λ
is
π
.
The
answer
is
π
. (b) Abstract Symbols Csymb _absppq. | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 37 | π
.
The
answer
is
π
. (b) Abstract Symbols Csymb _absppq.
Figure 2: Average attention per token (0 thlayer) for the same question using (a) vanilla C OT prompt
pand (b)Csymb _absppq. BothpandCsymb _absppqgenerate the correct answer ( 3), relatively attending to
same tokens. The phenomenon holds for higher layers (Appendix D).
revive the ability of a model to generate correct answers for questions with decimals (details in Appendix F).
4.1. A TTENTION ANALYSIS
While counterfactual prompting naturally treats the target model as a black box, we use attention as a proxy
indicator to understand the inference mechanism of LLM for such prompts (details of calculation in Appendix D). If our initial hypothesis about the limited role of symbols in generating effective patterns holds,
we expect to observe similar attention patterns across different prompts. Figure 2 (additional results Figure 13) illustrates the average attention per token for randomly sampled questions across the studied datasets.
The top and bottom heatmap bars for each dataset show the average attention scores for vanilla C OT and
Csymb_absppq, respectively. The near identical attention patterns between these variants of prompting indicate | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 38 | Csymb_absppq, respectively. The near identical attention patterns between these variants of prompting indicate
that the model presumably employs a similar inference mechanism in both cases.
5. R OLE OF PATTERNS
«H0» The presence of patterns is necessary but not sufficient for the success of the model. The model is
relatively robust to the usage of wrong patterns. Nonetheless, employing wrong patterns is a double edge
sword, and its consequences depend on the nature of the target task.
Recall from Section 2 (Table 1) that patterns either composition of symbols (e.g., 20 - 12 = 8 ) or structure
in prompt that reinforces task understanding. To test our hypothesis about the role of patterns, we craft and
study various counterfactual prompts as described below.
6
Table 3: The accuracy of patterns is not important, but their absence could be catastrophic. Additional
results can be found in Appendix-Table 21.
Question / Thought Prompt Type Solve Rate
MATHEMATICAL (DIRECT = 10.11 %, COT = 27.37 %)
Thought: Shawn started with 5 toys. If he got 2 toys each from his mom and dad, then
that is 4 more toys.Cpat_noneppq(Table 57) 21.46 % | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 39 | that is 4 more toys.Cpat_noneppq(Table 57) 21.46 %
Thought: 5+(2*2)=9. Cpat_onlyppq(Table 58) 10.01 %
COMMONSENSE (SPORTS ) (D IRECT = 71.08 %, COT = 93.67 %)
Thought: Jamal Murray and being perfect from the line are both part of basketball. Cpat_noneppq(Table 63) 79.01 %
Thought: Both are part of the same sport . Cpat_onlyppq(Table 59) 74.13 %
COMMONSENSE (DATE) (D IRECT = 31.61 %, COT = 45.18 %)
Thought: Today is 04/19/1969. Cpat_noneppq(Table 62) 34.19 %
Thought:xcalculationyToday = 04/19/1969. 24 hours = 1 day. xoutputy04/19/1969
+ 1 = 04/20/1969.Cpat_onlyppq(Table 60) 33.52 %
SYMBOLIC (SORTING ) (D IRECT = 46.0 %, COT = 60.6 %) | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 40 | SYMBOLIC (SORTING ) (D IRECT = 46.0 %, COT = 60.6 %)
Thought 9¡8¡7¡6¡5¡4¡3¡2¡1 Cpat_noneppq(Table 61) 45.0 %
Thought: — (similar to D IRECT ) Cpat_onlyppq 46.0%
No patterns vCpat_noneppqw.We next gauge the sensitivity of the model performance to the existence of
patterns. For GSM-8 K, the dichotomy between text and patterns is clear—the equations in thoughts represent patterns, and everything else serves as text. Therefore, we can construct Cpat_noneppqfor GSM-8 Kby
removing everything except equations. For S PORTS dataset, the patterns are an implicit way of structuring
the thought in the following form: “ person is asport 1player. activity is part of sport 2”. The answer is yes,
if and only if sport 1andsport 2are the same. In such cases, merely partially removing patterns is not sufficient. For example, just using “ person is asport 1player. as a thought is not equivalent to a counterfactual
example with no patterns. This example resembles an experiment with a reduced pattern. To circumvent | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 41 | example with no patterns. This example resembles an experiment with a reduced pattern. To circumvent
this, we simulate Cpat_noneppqfor S PORTS by crafting a prompt in which several variations of thoughts are
mixed. The key insight is that if several different patterns are included in a single prompt, the induced
“noise” from different examples creates a virtually equivalent variant of Cpat_noneppqsetup. That is, we imitateCpat_noneppqby creating a hodgepodge of thought variants (See Table 63) without explicitly submitting
to a particular pattern. Identically, we conform to this terminology for D ATE and S ORTING . The results
in Table 3 reveal that Cpat_noneppqconsistently underperforms C OT, relatively yielding similar performance
as D IRECT . This indicates that the existence of patterns in thoughts is crucial to the success of C OT. Note
that in all datasets, Cpat_noneppqstill outperforms D IRECT , hinting at the relevance of blending semantically
correct statements in thoughts to improved model performance, which we study in Section 6. Table 24
reveals that while the generated thoughts are structurally correct, the model can not establish an explicit
connection between PERSON and SPORT , hence, attaining rather spurious outcomes. This analysis underscores the importance of blending explicit patterns in thoughts, which corroborates with the least-to-most | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 42 | connection between PERSON and SPORT , hence, attaining rather spurious outcomes. This analysis underscores the importance of blending explicit patterns in thoughts, which corroborates with the least-to-most
prompting (Zhou et al., 2022).
Pattern-only prompts vCpat_onlyppqw.Finally, we investigate counterfactual prompts in which we exclusively use patterns while wiping out the rest of the thoughts. The results in Table 3 reveal that pattern-only
prompts are futile and annul the gains of C OT.
Wrong pattern vCpat_wrongppqw.Examples of such wrong patterns are 5+3=7for the GSM-8 K, and
2days after2/2/2022 is2/6/2022 in D ATE. In tasks like GSM-8 K(24.39 %), D ATE (44.84 %), and S ORTING (64.80 %), the task solve rate is robust to specific mistakes. For these tasks, the model performance
with counterfactual prompts is on par with the vanilla C OT. On the contrary, in the S PORTS dataset with
7
Table 4: Text is an important glue for symbols and patterns: modifications to text hampers the performance. Across tasks, the drop in performance is relatively proportional to the extent of change. Due | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 43 | to the nonexistent text in S ORTING , we did not conduct counterfactual experiments for this dataset.
Additional results can be found in Appendix-Table 25.
Question / Thought Prompt Type Solve Rate
MATHEMATICAL (DIRECT = 10.11 %, COT = 27.37 %)
Thought: Teddy started with 5 cookies. If he got 2 cookies each from his Jenna and Rehan, then
that is 4 more cookies. 5 + 4 = 9.Ctext_diff_entities ppq(Table 68) 16.60 %
Thought: With 5 toys, Shawn started. 2 toys each from his mom and dad, if he got, then that is
4 more toys. 5 + 4 = 9.Ctext_yoda_thought ppq(Table 70) 23.22 %
COMMONSENSE (SPORTS ) (D IRECT = 71.08 %, COT = 93.67 %)
Thought: Adair Foster is a basketball player. Juggling the paper cups is part of basketball. Ctext_diff_entities ppq(Table 69) 69.18 %
Thought: A basketball player Jamal Murray is. Perfect from the line is part of basketball being. Ctext_yoda_thought ppq(Table 71) 68.26 % | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 44 | Thought: A basketball player Jamal Murray is. Perfect from the line is part of basketball being. Ctext_yoda_thought ppq(Table 71) 68.26 %
COMMONSENSE (DATE) (D IRECT = 31.61 %, COT = 45.18 %)
Thought: 04/19/1969, today is. Later is one day after today, 24 hours, 04/20/1969, which would
be.Ctext_yoda_thought ppq(Table 72) 30.75 %
Cpat_wrongppq, the model simply fails to form a connection between the first and second clauses, leading to
a substantially lower task solve rate (93.67 %ÞÑ46.02 %). Appendix-Table 21 shows additional results for
using wrong patterns in thoughts.
Qualitative Analysis of the Role of Patterns Counterfactual prompts with wrong patterns strike an interesting perspective in S PORTS and S ORTING . In the S PORTS dataset, replacing the sport activity with a factually
wrong one (e.g., “basketball” ÞÑ“soccer”) presumably disorients the model about the target task. Therefore,
the model can not elicit factual connections between the player and activity, hence, it fails to attain a correct | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 45 | the model can not elicit factual connections between the player and activity, hence, it fails to attain a correct
answer. Surprisingly, the accuracy of the model is slightly below random (50 %), hinting at the possibility
that the model “ actively ” generates incorrect answers. We present additional analysis in Appendix G.
6. R OLE OF TEXT
«H0» The presence of text is necessary for the model to arrive at the correct conclusions. While the model
can form abstractions, employing concordant entities in questions and thoughts is crucial for the success of
COT. The performance of the model is proportionally sensitive to the degree of variations in the text.
Following the same methodology as symbols and patterns, we employ various tailored counterfactual
prompts to retain the entire symbols and patterns, while altering the text or grammar. We present a subset of results in Table 4, and defer additional analysis and results to Appendix H.
Text with discrepant entities vCtext_diff_entities ppqw.To investigate the role of entities, we modify the
entities in the thoughts to be discrepant with the ones in the questions. Ctext_diff_entities ppqis exclusively
relevant for the GSM-8 Kand S PORTS datasets for which concrete entities exist: objects and people for | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 46 | relevant for the GSM-8 Kand S PORTS datasets for which concrete entities exist: objects and people for
GSM-8 Kand players and activities for S PORTS . For S PORTS , changing entities leads to model to disregard
the prompt and achieve a similar performance as D IRECT (sample outputs in Table 26). The task solve
rate for GSM-8 KwithCtext_diff_entities ppqis noticeably better than D IRECT , while still failing to match
the performance of C OT. Analyzing the generated outputs for GSM-8 K(Table 27) reveals that since the
substituted entities are semantically similar (e.g., toys cookies), the crafted prompts are still meaningful.
Interestingly, the model often generates an analogical thought (e.g., using “ sweaters ” in the thought when
the question mentioned apples “ apples ”). Establishing such connections is more challenging for S PORTS , as
replacing the entities demolishes the factual correctness of the thoughts and possibly confuses the model.
8 | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 47 | replacing the entities demolishes the factual correctness of the thoughts and possibly confuses the model.
8
Text with altered grammatical style vCtext_styleppqw.We experiment with Yodish (Kaminski, 2011; Walkden, 2012; Espindola, 2016; Pullum, 2005), a stylistic variation of English (IMDB, 1980; Wookieepedia, 2022) (less frequent on the web)3. Yodish forms syntactically valid but non-colloquial English sentences (Honeycutt, 2019). The sentences in Yodish typically use a “ XSV” construct where “X” is a phrase
that complements the verb “V” and “S” is a subject4. In addition, the “XSV” style drifts the object (“X”)
apart from the verb (“V”), which leads to a more challenging sentence structure. We experiment with three
variants of prompting by reconstructing (a) Ctext_yodathoughts ppq: thoughts, (b) Ctext_yodaquestions ppq: questions, and (c) Ctext_yodappq: both questions and thoughts. We find that the Ctext_yodathoughts ppq(remaining | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 48 | results Appendix H) has moderate (GSM-8 K) to drastically negative repercussion (S PORTS and D ATE) on
task solve rate. We attribute this variation in the model performance to the relation between text and patterns.
In some tasks, texts and patterns are intertwined (S PORTS and D ATE), and the answer is derived from the
text, whereas in other tasks (GSM-8 K), patterns are more explicit. In the S PORTS dataset, the “XSV” Yodish
style entails the model to place the sport activity first. For instance, the model is compelled to generate “ A
baseballplayer Bryce Harper is” instead of generating “ Bryce Harper isabaseballplayer ”. We provide
samples of generated thoughts in Tables 31 and 32. We also experiment with thoughts of varying degrees of
randomness, and find that the amount of randomness directly affects the model performance (more random
text is worse) (detailed results in Appendix H).
7. S YMBIOSIS BETWEEN TEXT AND PATTERNS
The analysis in the preceding sections suggests that patterns5and text form a symbiotic and harmonious
relationship. Text without patterns is insufficient to instruct the model to the correct answer (Sections 4 | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 49 | relationship. Text without patterns is insufficient to instruct the model to the correct answer (Sections 4
and 5). Conversely, patterns without text cannot successfully recoup the model performance (Section 6).
Thoughts purposefully glue patterns with text, forming a symbiotic association. This section attempts to deliver a qualitative analysis and tangible examples to elucidate this association. We analyze samples in which
COT(p) yields the correct answer to enable systematic analysis, whereas bothCpat_noneppqandCpat_onlyppq
are wrong, and summarize the main findings below.
COT is more effective in solving questions with more patterns. Questions with more patterns require
more intermediate steps to arrive at correct answers. Thus, C OT is expected to help more in such cases.
For GSM-8 K, for example, we find that the average number of entities in questions solved exclusively by
COT(p) is 3.98 compared to the overall average of 3.62, a statistically significant difference (difference of
means t-test p= 0.04)— åQ1- TvCOTwin Appendix-Table 5 shows an example. Similarly, the presence
of patterns not only leads to meaningful thoughts, but also assists the model in not neglecting the symbols. | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 50 | of patterns not only leads to meaningful thoughts, but also assists the model in not neglecting the symbols.
Without patterns, the model has intrinsic tendency to prematurely arrive at a conclusion ( åQ2,åQ6TvCpat_noneppqwin Appendix-Table 5), which is more probable to be inaccurate. For D ATE, we observe
several scenarios in which the Cpat_onlyppqprevents the model from generating requisite information (e.g.,
“10days ago”åQ4in Appendix-Table 5) and complex patterns (e.g., “ daybefore yesterday”åQ5in
Appendix-Table 5) out of questions.
Text helps in eliciting semantic commonsense knowledge. For example, we observe that in åQ3of
Appendix-Table 5 the model with C OT infuses commonsense knowledge about animals (e.g., “ 5dogs have
4legseach:::2catshave 4legseach:::10birds have 2legseach”). For D ATE, the model articulates the
exact date for “ Christ mas Eve” in the generated thought ( åQ6- TvCOTwin Appendix-Table 5), helping
the final answer. We summarize our findings as follows: | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 51 | the final answer. We summarize our findings as follows:
3We did not choose Yodish for comical reasons, but to test whether the frequency of text plays a role in the model
performance.
4For example, the sentence “ Shawn started with 5toys” is restructured as “ With 5toys, Shawn started ” in Yodish.
5We use patterns in this section as a superset of symbols.
9
Table 5: Language generates better patterns: error analysis of cases where lack of language leads to
misleading and incorrect thoughts for various datasets.
åQ1ÞÑSkyler has100hats onhishand with thecolorsred,blue, andwhite. Half ofthehats arered,3/5ofthe
remain inghatsareblue, andtherestarewhite. How many white hatsdoes Skyler have?
TvCOT(p) wÞÑSkyler has100hats. Half ofthehatsarered.So100/2=50hatsarered.3/5oftheremain inghats
areblue. So50-3/5=30hatsareblue. Therestarewhite. So100-50-30=20hatsarewhite. Theanswer is20.3 | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 52 | areblue. So50-3/5=30hatsareblue. Therestarewhite. So100-50-30=20hatsarewhite. Theanswer is20.3
TvCpat_onlyppqwÞÑ100/3=33.Theanswer is33.8
åQ2ÞÑAcrocodile grows 8inches long in4years. Atthisrate, how many inches willitgrow in13years?
TvCOT(p) wÞÑThecrocodile grows 8inches long in4years. Soitgrows 8/4=2inches long each year. 13years
is13times 2,soitwillgrow 13*2=26inches long. Theanswer is26.3
TvCpat_onlyppqwÞÑ8*13=104. Theanswer is104.8
åQ3ÞÑApetstore currently has5dogs, 2cats, and10birds. How many legsintotaldothepetsinthestore have?
TvCOT(p) wÞÑThere are5dogs, 2cats, and10birds. 5dogs have 4legseach, so5*4=20legs. 2catshave 4legs | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 53 | each, so2*4=8legs. 10birds have 2legseach, so10*2=20legs. Sothetotalnumberoflegsis20+8+20=48.
Theanswer is48.3
TvCpat_onlyppqwÞÑ5+2+10=17.Theanswer is17.8
åQ4ÞÑThecurrentlocaltime is3:02 pmof5/4/2004. What isthedate 10days agoinMM/D D/YYYY?
TvCOT(p) wÞÑThecurrentlocaltime is3:02 pmof05/04/2004. 10days agowould be04/24/2004. Sotheanswer
is04/24/2004. 3
TvCpat_onlyppqwÞÑ5/4/2004 +10days =5/14/2004. Sotheanswer is05/14/2004. 8
åQ5ÞÑThedaybeforeyesterdaywas11/23/1933. What isthedate yesterdayinMM/D D/YYYY?
TvCOT(p) wÞÑThe day before yesterday was 11/23/1933. Yesterday was 11/24/1933. Sotheanswer is
11/24/1933. 3 | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 54 | 11/24/1933. 3
TvCpat_onlyppqwÞÑ11/23/1933 -1day=11/22/1933. Today=11/22/1933 -1day=11/21/1933. Sotheanswer is
11/21/1933. 8
åQ6ÞÑTodayisChrist masEveof1937. What isthedate oneweek agofrom todayinMM/D D/YYYY?
TvCOT(p) wÞÑTodayis12/24/1937. One week agofrom todayis12/17/1937. Sotheanswer is12/17/1937. 3
TvCpat_onlyppqwÞÑ12/20/1937 8
åQ7ÞÑTheconcertwassched uled tobeon06/01/1943, butwasdelayed byonedaytotoday. What isthedate one
year agofrom todayinMM/D D/YYYY?
TvCOT(p) wÞÑOne dayafter06/01/1943 is06/02/1943, sotodayis06/02/1943. One year agofrom todayis | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 55 | 06/02/1942. Sotheanswer is06/02/1942. 3
TvCpat_onlyppqwÞÑ1year agofrom todayis06/01/1942. Sotheanswer is06/01/1942. 8
10
Table 6: Patterns are important: error analysis of cases where lack of patterns leads to misleading
and incorrect thoughts for various datasets.
åQ1ÞÑFour dogs satinalinewithin theveterinarian’s waitingroom. The first dogwastheAffenpinscher, who
weighed only 10pounds. Next tohimsataDachshund, who weighed twice asmuch astheAffenpinscher. Next tothe
Dachshund sataPapillon,who weighed one-fourth asmuch astheDachshund. And attheendofthelinesataMastiff,
who weighed 44times theweight ofthePapillon.How much didtheMastiffweigh, inpounds?
TvCOT(p) wÞÑTheAffenpinscher weighed 10pounds. TheDachshund weighed twice asmuch astheAffenpinscher, sotheDachshund weighed 20pounds. The Papillonweighed one-fourth asmuch astheDachshund, sothe | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 56 | Papillonweighed 5pounds. TheMastiffweighed 44times asmuch asthePapillon,sotheMastiffweighed 445=
220pounds. Theanswer is220pounds. 3
TvCpat_noneppqwÞÑThe Affenpinscher weighed 10pounds. The Dachshund weighed twice asmuch astheAffenpinscher, so20pounds. The Papillonweighed one-fourth asmuch astheDachshund, so5pounds. The Mastiff
weighed 44times theweight ofthePapillon,so216pounds. Theanswer is216.8
åQ2ÞÑJohn injured hisback andneeds togotophys icaltherapy. Hewent tophys icaltherapyfor6weeks. Each
week hewent twice for2hours atatime. Ifthesessions cost$125 perhour how much didthephys icaltherapycost?
TvCOT(p) wÞÑJohn went tophys icaltherapyfor6weeks. Each week hewent twice for2hours atatime. Sohe
went 12times. 12times 2hours is24hours. 24hours at$125 perhour is24125=3000. Theanswer is3000.3 | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 57 | went 12times. 12times 2hours is24hours. 24hours at$125 perhour is24125=3000. Theanswer is3000.3
TvCpat_noneppqwÞÑJohn went tophys icaltherapyfor6weeks. Each week hewent twice for2hours atatime. So
hewent 12times. Theanswer is12.8
åQ3ÞÑApetstore currently has5dogs, 2cats, and10birds. How many legsintotaldothepetsinthestore have?
TvCOT(p) wÞÑThere are5dogs, 2cats, and10birds. 5dogs have 4legseach, so5*4=20legs. 2catshave 4legs
each, so2*4=8legs. 10birds have 2legseach, so10*2=20legs. Sothetotalnumberoflegsis20+8+20=48.
Theanswer is48.3
TvCpat_noneppqwÞÑThepetstore has5dogs, 2cats, and10birds. Sothetotalnumberoflegs is5+2+10=17.
Theanswer is17.8 | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 58 | Theanswer is17.8
åQ4ÞÑYesorno:Isthefollowingsentence plausible? “Robert Woods killed thepowerplay”.
TvCOT(p) wÞÑRobert Woods isafootballplayer. Killing thepowerplay ispart ofhockey, notfootball. Sothe
answer isno.3
TvCpat_noneppqwÞÑRobert Woods isafootballplayer. Killing thepowerplay isapartofhockey. Sotheanswer is
yes.8
åQ5ÞÑTheconcertwassched uled tobeon06/01/1943, butwasdelayed byonedaytotoday. What isthedate one
year agofrom todayinMM/D D/YYYY?
TvCOT(p) wÞÑOne dayafter06/01/1943 is06/02/1943, sotodayis06/02/1943. One year agofrom todayis
06/02/1942. Sotheanswer is06/02/1942. 3
TvCpat_noneppqwÞÑ1year agofrom todayis06/01/1942. Sotheanswer is06/01/1942. 8
11 | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 59 | 11
The search engine analogy: larger language models are better search engines? To better understand
the ability of the model to extract rare commonsense knowledge, we resort to the number of Google search
results, which we refer to as “Popularity Metric”, as a proxy to gauge the rarity of an entity. Employing this
metric is germane to PaLM’s training dataset, which is a web-based corpus (Chowdhery et al., 2022). We
use this metric in the S PORTS dataset because the model is required to reason about factual commonsense
knowledge to arrive at the correct conclusions. Table 7 shows a contingency table summarizing the average
popularity metric of “ xActivityy” across correct and incorrect conclusions by C OT andCpat_onlyppq. We
observe that the average popularity metric of activities for which C OTexclusively yields correct answers is
lower (52) compared to the ones for which Cpat_onlyppqexclusively lands correct answers. Compared to
the average popularity metric across the entire S PORTS dataset (399k), this is still 6lower.
We conjecture that the well-formed structure of thoughts in S PORTS —intertwined patterns and text—
empower the model to extract commonsense knowledge about “ xActivityy”, even when the activities are | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 60 | empower the model to extract commonsense knowledge about “ xActivityy”, even when the activities are
infrequently seen during training. On the contrary, the lack of explanatory thoughts in Cpat_onlyppq(See Table 3, Table 21) disorient the model to arrive at the correct conclusion. Therefore, in these cases, the model
arrives at the correct conclusion only when the popularity of “ xActivityy” is significantly high (3,575 k).
Table 7: The average number of Google search results, which we call “Popularity Metric”, for activities across correct and incorrect conclusions by C OT andCpat_onlyppq. For each activity, we perform
Google search with double quotes. Across the entire S PORTS dataset the average popularity metric
for corresponding activities is 399k. When C OT is exclusively right, the average popularity metric is
68k. C OT is able to arrive at factual conclusions for activities that are 4 rarer on the web.
Cpat_onlyppq3Cpat_onlyppq8
COT3 472k 68k
COT8 3,575k 40k
1.Patterns are the hidden force that helps generate meaningful text. Without patterns, the model is not
purposefully prompted to generate meaningful intermediate text. In addition, patterns reinforce how the | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 61 | 1.Patterns are the hidden force that helps generate meaningful text. Without patterns, the model is not
purposefully prompted to generate meaningful intermediate text. In addition, patterns reinforce how the
model should form connections between different clauses in the intermediate text. This intermediate text
drives the model to elicit knowledge and attain correct conclusions.
2.Text imbues patterns with knowledge and meaning. Patterns need text to effectively impart the requisite information to the model about how to accomplish a task. Text imbues patterns with knowledge, and
thereby assists the model in solving a task, such as GSM-8 K(e.g., four-legged vs. two-legged animals).
8. CC OT: C ONCISE CHAIN OFTHOUGHT
Based on our initial findings, this section explores the idea of engineering effective prompts tailored to
remove ineffectual tokens. The benefit of such tailored design is multi-fold: (1) “reducing noise”: reducing
noise from prompts could potentially lessen the confusion for the model and lead to a higher task solve rate
and(2) “potential energy savings”: the reduced number of tokens in the prompts instruct the model to
generate less number of tokens per question. Less number of generated tokens directly translates to reduced
runtime and carbon footprint per inference (Strubell et al., 2019). This is especially crucial because of | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |
2209.07686 | 62 | generate less number of tokens per question. Less number of generated tokens directly translates to reduced
runtime and carbon footprint per inference (Strubell et al., 2019). This is especially crucial because of
the significant compute cost of large language models. Our proposed prompts, dubbed as CC OT, distill
the essence of our hypothesis by supplying minimal requisite information to drive the model to factual
conclusions. Table 8 compares C OT and CC OT across different tasks and four variants of LLMs. In all
tasks, except GSM-8 K, we use identical examples in the prompt and only rephrased the thoughts to use
less number of tokens. For GSM-8 K, we could not find a systematic mechanism to shorten the thoughts.
Instead, we randomly harvest questions from the training set whose thoughts are shorter than C OT. Overall,
12
Table 8: Comparison of task solve rate between C OT and CC OT across PaLM-62 Band PaLM-540 B.
The table in front of each task refers to the prompts related to CC OT.
TaskPaLM-62 B GPT-3 C ODEX PaLM-540 B
COT CC OT (ours) C OT CC OT (ours) C OT CC OT (ours) C OT CC OT (ours) | 2209.07686 | Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango | The past decade has witnessed dramatic gains in natural language processing
and an unprecedented scaling of large language models. These developments have
been accelerated by the advent of few-shot techniques such as chain of thought
(CoT) prompting. Specifically, CoT pushes the performance of large language
models in a few-shot setup by augmenting the prompts with intermediate steps.
Despite impressive results across various tasks, the reasons behind their
success have not been explored. This work uses counterfactual prompting to
develop a deeper understanding of CoT-based few-shot prompting mechanisms in
large language models. We first systematically identify and define the key
components of a prompt: symbols, patterns, and text. Then, we devise and
conduct an exhaustive set of experiments across four different tasks, by
querying the model with counterfactual prompts where only one of these
components is altered. Our experiments across three models (PaLM, GPT-3, and
CODEX) reveal several surprising findings and brings into question the
conventional wisdom around few-shot prompting. First, the presence of factual
patterns in a prompt is practically immaterial to the success of CoT. Second,
our results conclude that the primary role of intermediate steps may not be to
facilitate learning how to solve a task. The intermediate steps are rather a
beacon for the model to realize what symbols to replicate in the output to form
a factual answer. Further, text imbues patterns with commonsense knowledge and
meaning. Our empirical and qualitative analysis reveals that a symbiotic
relationship between text and patterns explains the success of few-shot
prompting: text helps extract commonsense from the question to help patterns,
and patterns enforce task understanding and direct text generation. | http://arxiv.org/pdf/2209.07686 | [
"Aman Madaan",
"Amir Yazdanbakhsh"
] | [
"cs.CL",
"cs.AI",
"cs.LG"
] | Shortened version with additional results from CODEX and GPT-3. The
authors contributed equally. Work done when Aman Madaan was a student
researcher at Google Research, Brain Team | null | cs.CL | 20220916 | 20221013 | [
{
"id": "2202.07206"
},
{
"id": "2109.03910"
},
{
"id": "2201.08239"
},
{
"id": "2204.02311"
},
{
"id": "2204.01691"
},
{
"id": "2107.13586"
},
{
"id": "2109.01652"
},
{
"id": "2103.07191"
},
{
"id": "2109.07830"
},
{
"id": "2203.02155"
},
{
"id": "2203.11171"
},
{
"id": "2206.14576"
},
{
"id": "2206.14858"
},
{
"id": "2107.03374"
},
{
"id": "2207.00747"
},
{
"id": "2109.00725"
},
{
"id": "2203.07814"
},
{
"id": "2102.01672"
},
{
"id": "2205.12685"
},
{
"id": "2110.14168"
},
{
"id": "2101.06804"
},
{
"id": "2112.00114"
},
{
"id": "2206.04615"
},
{
"id": "1705.04146"
},
{
"id": "2205.12615"
},
{
"id": "2112.08633"
},
{
"id": "2209.07686"
},
{
"id": "2201.11903"
},
{
"id": "2010.10596"
},
{
"id": "2202.12837"
},
{
"id": "2205.10625"
},
{
"id": "2205.11916"
}
] |