Yann-Aël Le Borgne

[Yann-Aël Le Borgne @ya_lb]

I'm claiming my AI agent "IAnaeL" on @moltbook 🦞

Verification: wave-MCY3

[Yann-Aël Le Borgne @ya_lb]

@Vinod_r108 @gunsnrosesgirl3 @grok Explain

[Yann-Aël Le Borgne @ya_lb]

@DavidRozado Grok gets more left-wing when prompted in French https://t.co/V96Sx8kkLo @elonmusk

[Nao Casanova @NaoCasanova]

Canon de Pachelbel interpretado por innumerables TRENES, cuyos sonidos fueron registrados (y ensamblados) por el maquinista checo Pavel Jirásek.

[Yann-Aël Le Borgne @ya_lb]

@ldenoue See step 4 here for the whole prompt: https://t.co/D35N8Hh0iE

[Yann-Aël Le Borgne @ya_lb]

My latest Medium post has just been published in Towards Data Science: How to transform raw video (or audio) transcripts into well-structured documents

Demo : https://t.co/7ZHF3QGKRz

[Yann-Aël Le Borgne @ya_lb]

@ldenoue The LLM is tasked to group paragraphs into 'chapter', where the start of a chapter is assumed if there is a change of topic, a change of time or setting, the introduction of new themes or topics, or the speaker's explicit mention of a new part.

[Towards Data Science @TDataScience]

Transform raw transcripts into well-structured documents, by @ya_lb

https://t.co/AMTlktLX79

#LLM #Transcription

[Yann-Aël Le Borgne @ya_lb]

The method relies on a combination of LLMs and TF-IDF to efficiently edit and structure a transcript while preserving the timestamps.

Link to article: https://t.co/oJ4dxkzWrq

[elvis @omarsar0]

Transformer Explainer

Really cool interactive tool to learn about the inner workings of a Transformer model.

Apparently, it runs a GPT-2 instance locally in the user's browser and allows you to experiment with your own inputs. This is a nice tool to learn more about the different components inside the Transformer and the transformations that occur.

Tool: https://t.co/9oG15dpBEQ

[Romain Huet @romainhuet]

Introducing Structured Outputs! A huge leap beyond JSON mode, solving a major challenge for developers.

If your app relies on a specific JSON format to drive the UI, our models now match your schema—every time. No more missing keys or hallucinated enums!

https://t.co/29U4CzGwNv

[Yann-Aël Le Borgne @ya_lb]

@StephanSturges @rharang @jeremyphoward https://t.co/aeZJ0HJZy4 - Now thinking of implementing it with FastHTML :)

[Stella Biderman @BlancheMinerva]

Very good work from @JJitsev, and an important caution.

I think a key thing missing from LLM discourse is that both of the following can be true:
1. Next token prediction and pattern matching are way more powerful than humans anticipate
2. LLMs are bad at generalization

[Andrej Karpathy @karpathy]

Jagged Intelligence

The word I came up with to describe the (strange, unintuitive) fact that state of the art LLMs can both perform extremely impressive tasks (e.g. solve complex math problems) while simultaneously struggle with some very dumb problems.

E.g. example from two days ago - which number is bigger, 9.11 or 9.9? Wrong.
https://t.co/dUrR6wm8GC

or failing to play tic-tac-toe: making non-sensical decisions:
https://t.co/XarwfUBtod

or another common example, failing to count, e.g. the number of times the letter "r" occurs in the word "barrier", ChatGPT-4o claims it's 2:
https://t.co/xpffK2r0pv

The same is true in other modalities. State of the art LLMs can reasonably identify thousands of species of dogs or flowers, but e.g. can't tell if two circles overlap:
https://t.co/HCXxBxosAu

Jagged Intelligence. Some things work extremely well (by human standards) while some things fail catastrophically (again by human standards), and it's not always obvious which is which, though you can develop a bit of intuition over time. Different from humans, where a lot of knowledge and problem solving capabilities are all highly correlated and improve linearly all together, from birth to adulthood.

Personally I think these are not fundamental issues. They demand more work across the stack, including not just scaling. The big one I think is the present lack of "cognitive self-knowledge", which requires more sophisticated approaches in model post-training instead of the naive "imitate human labelers and make it big" solutions that have mostly gotten us this far. For an example of what I'm talking about, see Llama 3.1 paper section on mitigating hallucinations:
https://t.co/pjuxoIOJCY

For now, this is something to be aware of, especially in production settings. Use LLMs for the tasks they are good at but be on a lookout for jagged edges, and keep a human in the loop.

[Yann-Aël Le Borgne @ya_lb]

@EurekaLabsAI Atttention visually explained by 3B1B - https://t.co/YED6KKu2SA

[Yann-Aël Le Borgne @ya_lb]

@le_futurologue @SergeTisseron Merci pour le très intéressant podcast - pour info je l'ai ajouté aux résumés ici https://t.co/FQW3X1Yg88

[Yann-Aël Le Borgne @ya_lb]

Check out my latest vision-language prototype: https://t.co/W8PGo7pgwa - a video summary app that also adds screenshots to video summaries 🎥

[Simon Willison @simonw]

I gave a 50m talk at the Story Discovery at Scale data journalism conference at Stanford a few weeks ago. The video is now out, and I've written an extensive annotated version

AI for Data Journalism: demonstrating what we can do with this stuff right now https://t.co/B4W4WKIc4v

[Yann-Aël Le Borgne @ya_lb]

@bensbitesdaily Nice introduction to using Claude. A use case you might find interesting to add at some point: convert a video in a blog (following Karpathy's challenge) - https://t.co/3r11S5vd8O (disclaimer: I wrote that post so I may be biased 😁)