Really cool work on data-constrained pretraining! Key idea is to assign repeated data an effectiveness: how many fresh tokens would have produced the same validation loss? That lets them put repeated and fresh data on a common scale. 1/7

Parallel inference is inevitable. @adityagrover_ took the @Ai4Conferences stage to make the case: GPUs parallelized matrix multiplication. Transformers parallelized training. Diffusion parallelizes inference. Sequential token generation has a ceiling. Diffusion breaks it.

1/ Today, @james_y_zou, @pmphlt, @jurafsky, and I are releasing string2string Studio: an open-source, in-browser platform that brings alignment, comparison, search, and evaluation into one interactive, visual environment for strings and sequences. See:

So proud of my former student @marionlepert. After a melanoma scare of her own, she channeled it into building something that could help thousands of others: OpenDerm, an open-source robot for high-resolution skin imaging, aimed at catching melanoma earlier. All OpenSource 🙌

People often blame academics for all the failings of AI conferences, but 36% of the papers at NeurIPS+ICML+ICLR in 2024 had at least one industry author. Could we agree to at least share the blame proportionally between academia and industry?

Thanks for having us on @RoboPapers! It was great to discuss our work Chorus, an approach to coordinate multiple robots in a decentralized manner. Looking forward to the next episodes @chris_j_paxton @DJiafei @micoolcho

Why do Vision-Language-Action (VLA) models still struggle with contact-rich manipulation? 🤖 Excited to share our new paper: Demystifying When and Why VLAs Fail in Contact-Rich Tasks and How to Fix Them. 🧵 (1/8)

I had so much fun creating the Self-Improving AI Agents course with @achowdhery, and teaching it twice in one year at Stanford! We also collaborated with Stanford Online to make the course available online: YouTube: The field is moving incredibly fast, but we tried to focus on the core concepts that help us build better AI systems. I hope you enjoy it!

Combining real-time interactivity, task understanding, and full-body action prediction on a humanoid is so, so hard. Here's an example where we bring all of these together in Gemini Robotics 2 🤖🧠

Catching skin cancer early is a home robotics problem. Melanoma is highly treatable when detected early, yet today’s screening process depends heavily on patients noticing tiny changes across their entire skin surface. This requires patients to solve a near-impossible visual-memory and registration problem. I built OpenDerm, an open-source 4-DOF robot that captures high-resolution images of the skin and uses them to reconstruct and track the skin surface in 3D over time. The best way to make skin screening truly routine is to bring it into the home. OpenDerm shows that inexpensive robotic skin imaging is possible, but the path to scale is not a dedicated screening robot in every household—it is to make skin screening one of the many useful things a general-purpose home robot can do. Read more about why I built OpenDerm and how it works here: Blog: Project:

on a mission to make intelligence too cheap to meter 🚀🌖 @JonSaadFalcon and i had a blast talking about ipw and intelligence efficiency w/the 👸🏽 @JayaGup10 link to full podcast in comments below

What if you could test a drug on a digital twin of your own cells before taking a prescription? Stanford HAI Faculty Affiliate Emma Lundberg is leading a team building an AI foundation model to make that possible:

A thoughtful letter highlighting the critical importance of open frontier models — a must-read for researchers, developers, and anyone shaping the future of AI. We built the @nvidia #Alpamayo# open platform ( around the same philosophy: sharing state-of-the-art #Physical# #AI# models, data, and tools with the community to accelerate innovation and advance the development of safer, more capable #autonomous# #vehicles# and #Physical# #AI# systems.

We can instantly build knowledge into a Transformer block, no gradient descent required! New work w/ amazing team @jerrywliu, @ronnygjunkins, @EyubogluSabri, Atri Rudra and @HazyResearch! To learn more, checkout our: 📝Blogpost: 📄Paper: