Datasets:

didierlopes

/

my-tweets-dataset

like 0

ok this is sweet - adding widgets to the dashboard + context directly from copilot very happy about this

working on enabling @ to add widgets to context from the chatting window directly

@davecraige @openbb_finance It's free for individuals. here's the open source code that was used to extend the data and add a custom agent on OpenBB

The cost of building software is plummeting to zero. 🤯 AI is transforming the game, making custom solutions easier and faster to develop than ever before. But here's the twist: this shift is fundamentally changing where the value lies. Desktop financial applications? Their…

@nikunj Geo and background is also a big advantage here I ended up stumbling upon the world of startups with an open source project that went viral, but prior to that I didn't know you could just start a company - in my head you needed to have a lot of money to do that I do think that…

@BrandonWatson FOR REAL 😂😂😂

@nikunj Good post. I also think that starting a startup at a young age is such an advantage. - it gives you exposure to hard things very early on - you understand quickly what you like and don't like to do - it gives you perspective in terms of speed of execution compared to a big org…

@fiege_max at least the person who reached out, the clients idk 😂 but now I am curious about who pays for professional matchmaking - giving me these vibes

@mattmaximo1 the only viable option this is how i'm gonna do b2b sales going forward

@fiege_max Nop, E

there's just no way

Trump shopping be like

@VijarKohli this is part of our custom AI agent framework that users can build - Theo just went overboard and built a strong example of what is possible :)

i believe this is my first open source project fully in typescript. thank you cursor.

this is one of the best demos of a custom AI agent I've seen on OpenBB

Excited to try this one out 🔥🔥🔥

Excited to share a mini-tool I built to keep writers accountable. As developers, we track our code commits. But what about tracking our writing progress? I created an open-source blog tracker that: - Generates beautiful SVG visualizations of your blog history - Supports both…

Jordi is a rockstar. You should definitely check this out.

Accessing a school system and it took over 30 seconds to update my password. Pretty sure this system is still the same it was 10y+ ago.

@theo this is so true or coding doesn't feel as much fun

Finally someone said it. The future of programming isn’t about giving up control to chatbots - it’s about amplifying the programmer’s agency. Cursor gets it. Faster iteration, granular control, and tools that adapt to YOU. It's no surprise that Cursor, for me, is the best…

@zaidmukaddam Amazing! Congrats 🙏

@mattmaximo1 but anything goes as long as price goes up, right?

@mattmaximo1 this level of gaslighting is wild 😂

@nickbae91 @JosephJacks_ any feedback let me know!

@gillinghammer Yup. With the right access to tools: - SEC filings - News - Grok for sentiment I think this agent could actually be pretty good at detecting bs company fundamentals At least that's what I had in mind. It also highlights why the agent would benefit from being integrated within…

btw it was incredibly simple to create the agent - all you need to do is in this video. The main thing is scraping X data from the account you want to re-create. Then I just used AI to update the character profile based on the profile I wanted to recreate, and it worked…

Next step is to integrate it within OpenBB workspace This would enable the agent to have access to all the data accessible within OpenBB along with: - context management - citations - artifacts generation - unstructured and structured files upload - access to the web - etc...

Created an Hindenburg Research agent utilizing Eliza framework.

Playing with Eliza agentic framework today from @ai16zdao

Fully open source repo can be found here:

Step-by-step guide can be found here:

Over the Christmas break, I decided to explore the world of fine-tuning while assessing the quality of open-source models that can run locally. This exploration was particularly important for me, as we frequently discuss with prospects the possibility of integrating local AI…

@kellyann3644 @amanrsanger @sualehasif996 @deepseek_ai I had to double check myself that it happened last week ahahah

@kellyann3644 @amanrsanger @sualehasif996 @deepseek_ai DeepSeek-v3 is!

This was an amazing podcast from @cursor_ai founders with a ton of alpha for teams building with AI. It also made it clear why Cursor will stay ahead of the game for a long time - they are obsessed over UX and iteration. Bullish on Cursor.

The DeepSeek-V3 paper highlights the adoption of MLA

DeepSeek-V3 (released LAST WEEK) achieved a significant breakthrough in inference speed over previous models. It tops the leaderboard among open-source models and rivals the most advanced closed-source models globally.

MLA architecture was first introduced by DeepSeek in May 2024.

this is insane - @amanrsanger and @sualehasif996 talking about @deepseek_ai Multi-head Latent Attention (MLA) architecture 2 months ago.

@rootquant If you need help integrating data, let's setup a channel w our team so we can help you out!

Beautiful dashboard!

@leeerob Pretty good

@0xwitchy What else do you think is missing?

@0xwitchy While I agree w most said, I think these AI agents are missing a native interface. You mention plugins for actions, which is great. But these tend to be third party APIs where the agent has limited control over what it can pull/push. Imo they will require a native AI interface…

@MuddyRC @_kaitoai Great post!

If you are in your early 20s, listen to this podcast Go into debt if you have to

Goal for 2025 - sleeping 8h, or at least trying to be in bed before 11pm. Let's see how long I last. My avg bed time is 2ish am.

Starting the year with something that we've been working on for the past few weeks: > Improving the experience that developers have in building their own applications on OpenBB. In this 5min video, I share how I am able to add a table and a chart quickly to OpenBB. All code is…

@pratty_agi @jmaslek11 😂

This is a great post by Vitalik on "From prediction markets to info finance"

@rootquant There's a few here that I would highly recommend. For instance, I really enjoyed The Art of Learning, When kale and coffee compete and Amp it up.

My EB2-EA petition was sent out today. 🤞🇺🇸

using the other app to play with agents

@tinztwins Happy new year!!!

I read 31 books in 2024. In 2025 I will be reading less. I will be consuming more podcasts & y2b on content I'm curious about. Also, I will be spending more time going deeper into AI & finance topics. During the holidays I had a lot of fun understanding better how fine-tuning…

Happy new year everyone 🙏

@yangWao 🙏🙌

@amadad @sgzsh269 @FireworksAI_HQ 👀👀👀👀👀

@KemarTiti an investment management firm!

@trac3abl3 I don't doubt in our vision for a second. But we spend all of our capital on the product, and sometimes we could spend more on marketing. So being recognized for something that we've been working on for 3+ years is nice :)

So this just happened

saw this over the weekend - i forgot how good this movie was

@0xSamHogan Yes

@HanchungLee Ya, I thought so too. Thought it needed to be a much higher number, to produce that type of output. But then again this is a probability game - so maybe you are just lucky, or unlucky - depending on your pov

@HanchungLee isn't this expected as with T>1 the softmax tends to result in a more uniform distribution - hence more likely to pick gibberish tokens

@bovaird_zach I was able in the end to fine-tune the phi-3-mini-4k-instruct model using MLX. I have M3. But results aren't the best. Need to spend more time on it to understand what's going on.

@pratty_agi @AlexReibman Yup. I remember us talking with @teoliphant about this a couple years back, and about the entire python dependency management ecosystem.

Creating a Telegram bot is SO ridiculously easy. Definitely built by developers who understand that any friction will result in huge drop offs.

@pratty_agi @AlexReibman Yup - but the alternatives are worse for larger projects. It's the difference between maintaining this poetry toml file as opposed to this pip requirements-full.txt (in this case, i believe requirements.txt was actually created…

@AlexReibman I do get that to be fair. I don't think you can get away from not using poetry for really large projects with a lot of python dependencies. It can get really messy - e.g. but for most projects you don't need it.

@AlexReibman The reason I haven't yet done this yet is because I've been using poetry for a long time. I like how I can lock dependencies - but I know many people think is overkill for simpler projects

@bovaird_zach Fine-tuning on a QA dataset that was extracted from my blog I wanted to see how a fine-tuned model would compare with answers that I would give - based on data sampled from the same distribution to the one that was used for training. And then I want to…

AI if there were no red teams

@dionizije_fa I want to do both :) but was just experimenting fine-tuning

the fact that we can just open the laptop and fine tune an AI model to wtv we want in a couple of hours from home is mind blowing to me

@souzatharsis This is amazing content, thank you so much for sharing!!! Will go through it and might look into getting the book when it comes out :) Also noticed that you used DocLing to convert PDF to markdown - and from my testing, it's the most accurate for PDFs with tables and text.…

When comparing the fine-tune model vs the base model, the results are pretty much identical. I'm evaluating it using Llama 3.2. as a judge since output is qualitative

This is the training loss error I get

not super happy with the results yet - but fine-tuning on a relatively small dataset. Should maybe look into fine-tuning with another model? any suggestions?

@yb_effect Good thread!

@VijarKohli @openbb_finance Thank you Vijar! 🙏🙏🙏

Since our launch on October 7th, we realized that while the majority of our audience is based in the US - we have users utilizing @openbb_finance from all around the globe. In fact, if we count sign ups since October we have a 84% country representation. Today, most of the top…

@ferrants Yes!

I turn this on when I'm working on OpenBB stuff for copy - e.g. marketing website. So that OAI models have more info on OpenBB in their weights

@skeptrune Nice! Just found instructions - I don't think I will need it, but if I do I will ping you. 🫡

@skeptrune This one is working for me At least for the first couple reports it worked perfectly, for others there were a few artifacts but I'm currently able to overcome most of these by doing structured output with gpt-4o

@skeptrune need to run it offline :/

@rickwong888 @sgzsh269 Yup!

@_Repete_ Happy holidays!!

@waydegilliam Using an LLM to do OCR is great if it's a one-off or you don't know what to expect from the document. When you know the format of the reports you are processing and have enough of them that can cover the edge cases - then reliability is important

@HakinaMadey @jxnlco I'm looking to run it offline!

@mattmaximo1 Yessiir

@azzabazazz Come to NYC on a Saturday and we can play together 🫡

Last time I played soccer with snow I was maybe like 10 - this should be fun

this profile summaries is actually pretty good it would be cool to have a signal from 0 to 10 on how likely you would be to enjoy the content of a person based on following and people you engage w the most

@dzhng @sgzsh269 I wanted to try llama parse, but given it has sensitive data from client I can only run it locally

@sgzsh269 ARE U KIDDING ME Holy shit this is insane It literally worked seamleslly on the first try, you just saved me so many hours - thank you

what's best library to parse a PDF that has tables, footnotes, headers, superscript, etc... tried markitdown from msft but am not impressed

@yangWao @insightpartners @openbb_finance 😄