It is only rarely that, after reading a research paper, I feel like giving the authors a standing ovation. But I felt that way after finishing Direct Preference Optimization (DPO) by Rafael Rafailov @ NeurIPS Archit Sharma Eric Stefano Ermon Christopher Manning and Chelsea Finn. This

Envie de créer la Journée Mondiale du Flux RSS pour sensibiliser sur cette espèce menacée d'extinction. On pourrait en profiter pour rappeler leur existence et leur importance pour l'équilibre écosystémique du Web.

Im seeing unpleasant things

Data contamination is a huge problem for LLM evals right now. At Scale, we created a new test set for GSM8k *from scratch* to measure overfitting and found evidence that some models (most notably Mistral and Phi) do substantially worse on this new test set compared to GSM8k.

I'm really excited about these results for many reasons, but the most important is that we're starting to connect mechanistic interpretability to questions about the safety of large language models.

Another example of loaded jargon for LLMs. This should be the poster child of that. We should only be saying the first thing. Talk in plain language

Awesome and highly useful: FineWeb-Edu 📚👏 High quality LLM dataset filtering the original 15 trillion FineWeb tokens to 1.3 trillion of the highest (educational) quality, as judged by a Llama 3 70B. +A highly detailed paper. Turns out that LLMs learn a lot better and faster

Woah what??? Microsoft just dropped Florence-2 on Hugging Face with an MIT license!! Pretty huge. Florence was initially Microsoft’s internal CLIP model, and they now expanded it to do various tasks like captioning, object detection, OCR, … just by prompting the model

Happy to see our WP w Shogo Sakabe and David Weinstein (so many years in the making!) out. We examine the role of codifying knowledge in the spread of the Industrial Revolution. A little thread. 1/N

This is amazing. I have never seen a theory paper in physics being retracted, no matter how many mistakes. If they go through with this, it could have a big impact on the field. x.com/WPMcB1997/stat…

Steve McCormick It is fairly rare that you can point at an equation and say "this is obviously wrong" like in this case. More often you know they're wrong because an assumption they made disagrees with established results. Eg, I remember a case 15 years ago or so when someone repeated a

It's a bit sad and confusing that LLMs ("Large Language Models") have little to do with language; It's just historical. They are highly general purpose technology for statistical modeling of token streams. A better name would be Autoregressive Transformers or something. They

Erik Bernhardsson It's surprising how bad UX most CI providers have.

Multivac, how can the net amount of entropy of the universe be decreased? I apologize, but as an AI language model I am not able to answer, as reversing entropy is a highly complex, multi-faceted problem. Here is a nuanced look at how leading experts have approached the topic:

BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2024 #NobelPrize in Physics to John J. Hopfield and Geoffrey E. Hinton “for foundational discoveries and inventions that enable machine learning with artificial neural networks.”

What if you could make physics diagrams come alive? At #UIST2024, we will be presenting our paper, Augmented Physics, an ML-Integrated Authoring Tool for Creating Interactive Physics Simulations from Static Diagrams Co-authors: Yi Wen Nandi Zhang Jarin Rubaiat Habib Ryo Suzuki

🚀 Introducing DeepSeek-V3! Biggest leap forward yet: ⚡ 60 tokens/second (3x faster than V2!) 💪 Enhanced capabilities 🛠 API compatibility intact 🌍 Fully open-source models & papers 🐋 1/n

Unpopular opinion: benchmarks like these are moving the field in the wrong direction No I don't want an AI to be able to memorize (useless?) questions like "How many paired tendons are supported by a sesamoid bone?" in its weights I want the "intern", as Andrej Karpathy is suggesting

I'm organizing the 🌟 2025 Vibe Coding Game Jam Deadline to enter: 25 March 2025, so you have 7 days - anyone can enter with their game - at least 80% code has to be written by AI - game has to be accessible on web without any login or signup and free-to-play (preferrably its

BREAKING: MIT just completed the first brain scan study of ChatGPT users & the results are terrifying. Turns out, AI isn't making us more productive. It's making us cognitively bankrupt. Here's what 4 months of data revealed: (hint: we've been measuring productivity all wrong)