Bullish, in the coming decades majority of compute will be spent on ai for science

🚀 Thrilled to share that I’ve joined Periodic Labs in my dream role! We’re building end-to-end AI systems that act as scientists in the real world. Most people talk about “AI scientists” as LLMs that reason about data. We’re aiming higher...

"One of our goals is to discover superconductors that work at higher temperatures than today's materials" I'm happy to see William Fedus and team working toward breakthroughs like these! With AI starting to meaningfully contribute to scientific discover, I think it's the right time

I asked Liam: Why name our startup Periodic Labs? He said think about *periods* in time. And then it hit me: we define entire periods of history by their critical materials: Copper Age → Bronze Age → Iron Age → Silicon Age. The name says all about our mission: to discover

The team Periodic Labs is absolutely cracked. Favorite team I've ever worked with. We're building AI scientists, and our progress is breakneck - already in the physical lab! The world doesn't need another social media app. We're turning bits into atoms.

congrats, liam!

I joined Periodic Labs in May. We’re building AI scientists + autonomous labs to create breakthroughs you can hold. Longer take: xiangfu.co/science More on Periodic: periodic.com

if you want to do RL, you need to sample a wide variety of good solutions from your base model how do we get high variance, highly knowledgeable base models? midtraining. Periodic Labs seeks an expert researcher to midtrain Autonomous Scientists

Thrilled about the launch of Periodic Labs , where I’m involved part-time alongside my PhD! I’m excited to help build a full-fledged AI scientist and extremely grateful to work with such a talented team Ekin Dogus Cubuk William Fedus Xander Dunn Rohan Pandey among others!

.Slater Stich's Bain Capital Ventures article on Periodic is incredibly well written.

I used to get NVIDIA RTX 3060 Ti. Now I can be Costa Huang Ti 😀

Why haven’t LLMs done new science yet? One reason from Periodic Labs cofounder Ekin Dogus Cubuk: science is iterative. LLMs haven’t been able to iterate experimentally yet. “ LLMs can be very smart, but if they're not iterating on science, they won't discover science. To be honest,

periodic ❤️ open-source! for example, we’ve been collaborating with the PyTorch team to build the highest MFU gpt-oss training implementation (includes thinky sinky flexattn) here’s a few SFT runs of gpt-oss-20b & 120b, where i get ~24% MFU for 20b and ~8% for 120b

wow, never thought the project i'm fortunate enough to be working on would find it's way into a video from one of my favorite YouTubers: AI Explained! our work is highlighted as a more interesting use case for AI than video generation (Sora 3) no less! good intro to what we're up

Shoutout to vLLM team for developing a clean, infinitely hackable inference engine! vLLM V1 architecture is great for in-flight weight updates. Just don't stop vLLM, send POST /collective_rpc(method_name=update_weights), and watch inference continue with current KVs!

Periodic Labs is following the natural progression of reward functions since 2022 from human-preference models (in RLHF), from math + code verifiable RMs (in reasoning models), to reward functions grounded in the physical world (in end-to-end optimization against experiment)

we don't deserve Charlie Marsh

yo has anyone heard of this Olmo model, loss looks good

Cool work!