INDIA BECAME OLYMPIAD CHAMPIONS FOR THE FIRST TIME EVER. 🇮🇳 - They did "Rohit Sharma walk" while receving the Trophy.

In new work led by gokul kannan with Peter Kim, we show that protein language models learn allosteric interactions without any explicit supervision, evidence that evaluating protein LMs solely on their ability to learn 3D structure may lose important functional information.

TL;DR: LLMs don’t reason.

And then there was one

Protenix: Protein + X | ByteDance - A trainable PyTorch reproduction of AlphaFold 3 GitHub: github.com/bytedance/Prot…

DHONI + Ashwin + Jadeja....!!!! - It's time for rank turners in Chepauk.

“Hey Aravind Srinivas Perplexity would you guys consider presenting results in a Reddit-style format, where each ‘comment’ comes from an LLM predisposed to different ideological perspectives (liberal, conservative, etc.). This would better mimic real discourse and help users

Diffusion Without Tears is our attempt to make the score-matching + SDE interpretation of diffusion geometrically intuitive. If you're interested in our upcoming interview with Yang Song, I recommend reading this first! Link below.

we've been working on democratizing fast kernel writing on the PyTorch team. try the challenge, either you or your AI!

Inductive Moment Matching Luma AI introduces a new class of generative models for one- or few-step sampling with a single-stage training procedure. Surpasses diffusion models on ImageNet-256×256 with 1.99 FID using only 8 inference steps and achieves state-of-the-art 2-step

Flow matching has gained popularity recently which is better, diffusion or flow matching? They are formally equivalent Our purpose is to help practitioners understand and use these frameworks interchangeably -- **regardless of what it’s called**

Introducing Open Molecules 25, a foundational quantum chemistry dataset including >100M DFT calculations across 83M unique molecules, built with 6B core hours of compute! What does this mean for drug discovery, biology, and BioML? 1/

It is trivial to explain why a LLM can never ever be conscious or intelligent. Utterly trivial. It goes like this - LLMs have zero causal power. Zero agency. Zero internal monologue. Zero abstracting ability. Zero understanding of the world. They are tools for conscious beings.

That moment when World Champion Gukesh D won his game against World no.1 Magnus Carlsen! Video: Aditya Sur Roy/ ChessBase India #chess #chessbaseindia #norwaychess #gukesh

I always found it puzzling how language models learn so much from next-token prediction, while video models learn so little from next frame prediction. Maybe it's because LLMs are actually brain scanners in disguise. Idle musings in my new blog post: sergeylevine.substack.com/p/language-mod…

Announcing Ambient Diffusion Omni — a framework that uses synthetic, low-quality, and out-of-distribution data to improve diffusion models. State-of-the-art ImageNet performance. A strong text-to-image results in just 2 days on 8 GPUs. Filtering ❌ Clever data use ✅

A good denoiser learns the geometry of the image manifold. Thus it makes perfect sense to use denoisers to regularize ill-posed problems. This was a key reason we proposed Regularization by Denoising (RED) in 2016. A modest landmark in citations, but big impact in practice 1/4

"we fine-tune the last layer"

Diffusion models have analytical solutions, but they involve sums over the entire training set, and they don't generalise at all. They are mainly useful to help us understand how practical diffusion models generalise. Nice blog + code by Raymond Fan: rfangit.github.io/blog/2025/opti…

🚨 "Mathematical Foundations of Geometric Deep Learning", co-authored with Michael Bronstein 📚 Read the paper here: arxiv.org/abs/2508.02723 🧠 We review the mathematical background necessary for studying Geometric Deep Learning. #GDL #mathematics #deeplearning #AI