zackproser/combined_writing · Datasets at Hugging Face

text stringlengths 0 1.36k
Around March of 2023, Generative AI caused astonishment and existential dread for many creatives, allowing anyone with an internet connection to instantly conjure working code in any language, as well as images of any style, translations, and more, simply by asking for them in natural human language.

For example, you can ask a Generative AI tool:

> Write me a JavaScript class representing a typical saas app user, and include methods for updating and deleting the user's profile.

To which a Large Language Model (LLM) such as ChatGPT would instantly respond:

> To represent a typical SaaS app user in JavaScript, we can define a class named User with properties that might include an ID, name, email, and any other relevant information. For updating and deleting the user's profile, we can include methods within the class to handle these actions. Here's a sample implementation:

The Large Language Model would continue, writing the following code out to the user's screen in seconds:

Prior to the widespread availability of Generative AI tools, you more or less needed to understand Javascript, its most recent syntax changes, object oriented programming conventions and database abstractions, at a minimum, to produce this code.

You also needed to have recently gotten some sleep, be more or less hydrated and have already had your caffeine to create this simple example. And even the most highly-skilled keyboard-driven developers would have taken a bit longer than a few seconds to write this out.

## GenAI is not just for text or code...

Here's an example of me asking ChatGPT4 to generate me an image with the following prompt:

> I'm creating a video game about horses enhanced with Jetpacks. Please generate me a beautiful, cheerful and friendly sprite of a horse with a jetpack strapped onto its back that would be suitable for use in my HTML5 game. Use a bright, cheery and professional retro pixel-art style.

I can use Large Language Models (LLMs) like ChatGPT to generate pixel art and other assets for my web and gaming projects.

Within a few moments, I got back a workable image that was more or less on the money given my prompt.

I didn't have to open my image editor, spend hours tweaking pixels using specialized tools, or hit up my designer or digital artist friends for assistance.

## How does GenAI work?

Generative AI works by "learning" from massive datasets, to draw out similarities and "features"

Generative AI systems learn from vast datasets to build a model that allows them to produce new outputs.

For example, by learning from millions of images and captions, an AI can generate brand new photographic images based on text descriptions provided to it.

The key technique that makes this possible involves training machine learning models using deep neural networks that can recognize complex patterns.

Imagine you have a very smart robot that you want to teach to understand and use human language. To do this, you give the robot a huge pile of books, articles, and conversations to read over and over again.

Each time the robot goes through all this information, it's like it's completing a grade in school, learning a little more about how words fit together and how they can be used to express ideas.

In each "grade" or cycle, the robot pays attention to what it got right and what it got wrong, trying to improve. Think of it like learning to play a video game or a sport; the more you practice, the better you get. The robot is doing something similar with language, trying to get better at understanding and generating it each time it goes through all the information.

This process of going through the information, learning from mistakes, and trying again is repeated many times, just like going through many grades in school. For a model as capable as ChatGPT 4, the cost to perform this training can exceed $100 million, as OpenAI's Sam Altman has shared.

With each "generation" of learning, the robot gets smarter and better at using language, much like how you get smarter and learn more as you move up in school grades.

## Why is GenAI having its moment right now?

GenAI is the confluence of many complimentary components and approaches reaching maturity at the same time

**Data

**: Availability of massive datasets and computing power to train extremely robust models, some with billions of parameters

**Advanced architectures

**: New architectures like transformers that are very effective for language and generation

**Progressive advancement of the state of the art

**: Progressive improvements across computer vision, natural language processing, and AI in general

## Why is GenAI such a big deal?

Prior to the proliferation of LLMs and Generative AI models, you needed to have some pixel art skills, and be proficient in use of photo editing / creation software such as Photoshop, Illustrator, or GIMP in order to produce high quality pixel art.

Prior to Gen AI, you needed to be a software developer to produce working code.

Prior to Gen AI, you needed to be a visual artist to produce images, or a digital artist to produce pixel art, video game assets, logos, etc.

With Generative AI on the scene, this is no longer strictly true.

You do still need to be a specialist to understand the outputs and have the capability to explain them. In the case of software development, you still require expertise in how computers work, architecture and good engineering practices to employ the generated outputs to good effect.

There are some major caveats to understand around this piece such as why Generative AI is currently a huge boon to senior and above level developers, but commonly misleading and actively harmful to junior developers, but in general it holds true:

Generative AI lowers the barrier for people to produce specialized digital outputs.

[MIT News: Explained

Generative AI](https://news.mit.edu/2023/explained

generative

ai

1109)

[McKinsey

The State of AI in 2023: Generative AI's breakout year](https://www.mckinsey.com/capabilities/quantumblack/our

insights/the

state

of

Around March of 2023, Generative AI caused astonishment and existential dread for many creatives, allowing anyone with an internet connection to instantly conjure working code in any language, as well as images of any style, translations, and more, simply by asking for them in natural human language.

For example, you can ask a Generative AI tool:

> Write me a JavaScript class representing a typical saas app user, and include methods for updating and deleting the user's profile.

To which a Large Language Model (LLM) such as ChatGPT would instantly respond:

> To represent a typical SaaS app user in JavaScript, we can define a class named User with properties that might include an ID, name, email, and any other relevant information. For updating and deleting the user's profile, we can include methods within the class to handle these actions. Here's a sample implementation:

The Large Language Model would continue, writing the following code out to the user's screen in seconds:

Prior to the widespread availability of Generative AI tools, you more or less needed to understand Javascript, its most recent syntax changes, object oriented programming conventions and database abstractions, at a minimum, to produce this code.

You also needed to have recently gotten some sleep, be more or less hydrated and have already had your caffeine to create this simple example. And even the most highly-skilled keyboard-driven developers would have taken a bit longer than a few seconds to write this out.

## GenAI is not just for text or code...

Here's an example of me asking ChatGPT4 to generate me an image with the following prompt:

> I'm creating a video game about horses enhanced with Jetpacks. Please generate me a beautiful, cheerful and friendly sprite of a horse with a jetpack strapped onto its back that would be suitable for use in my HTML5 game. Use a bright, cheery and professional retro pixel-art style.

I can use Large Language Models (LLMs) like ChatGPT to generate pixel art and other assets for my web and gaming projects.

Within a few moments, I got back a workable image that was more or less on the money given my prompt.

I didn't have to open my image editor, spend hours tweaking pixels using specialized tools, or hit up my designer or digital artist friends for assistance.

## How does GenAI work?

Generative AI works by "learning" from massive datasets, to draw out similarities and "features"

Generative AI systems learn from vast datasets to build a model that allows them to produce new outputs.

For example, by learning from millions of images and captions, an AI can generate brand new photographic images based on text descriptions provided to it.

The key technique that makes this possible involves training machine learning models using deep neural networks that can recognize complex patterns.

Imagine you have a very smart robot that you want to teach to understand and use human language. To do this, you give the robot a huge pile of books, articles, and conversations to read over and over again.

Each time the robot goes through all this information, it's like it's completing a grade in school, learning a little more about how words fit together and how they can be used to express ideas.

In each "grade" or cycle, the robot pays attention to what it got right and what it got wrong, trying to improve. Think of it like learning to play a video game or a sport; the more you practice, the better you get. The robot is doing something similar with language, trying to get better at understanding and generating it each time it goes through all the information.

This process of going through the information, learning from mistakes, and trying again is repeated many times, just like going through many grades in school. For a model as capable as ChatGPT 4, the cost to perform this training can exceed $100 million, as OpenAI's Sam Altman has shared.

With each "generation" of learning, the robot gets smarter and better at using language, much like how you get smarter and learn more as you move up in school grades.

## Why is GenAI having its moment right now?

GenAI is the confluence of many complimentary components and approaches reaching maturity at the same time

**Data

**: Availability of massive datasets and computing power to train extremely robust models, some with billions of parameters

**Advanced architectures

**: New architectures like transformers that are very effective for language and generation

**Progressive advancement of the state of the art

**: Progressive improvements across computer vision, natural language processing, and AI in general

## Why is GenAI such a big deal?

Prior to the proliferation of LLMs and Generative AI models, you needed to have some pixel art skills, and be proficient in use of photo editing / creation software such as Photoshop, Illustrator, or GIMP in order to produce high quality pixel art.

Prior to Gen AI, you needed to be a software developer to produce working code.

Prior to Gen AI, you needed to be a visual artist to produce images, or a digital artist to produce pixel art, video game assets, logos, etc.

With Generative AI on the scene, this is no longer strictly true.

You **do** still need to be a specialist to **understand** the outputs and have the capability to **explain** them. In the case of software development, you still require expertise in how computers work, architecture and good engineering practices to employ the generated outputs to good effect.

**There are some major caveats to understand around this piece** such as why Generative AI is currently a huge boon to senior and above level developers, but commonly misleading and actively harmful to junior developers, but in general it holds true:

Generative AI lowers the barrier for people to produce specialized digital outputs.

[MIT News: Explained

Generative AI](https://news.mit.edu/2023/explained

generative

ai

1109)

[McKinsey

The State of AI in 2023: Generative AI's breakout year](https://www.mckinsey.com/capabilities/quantumblack/our

insights/the

state

of