We’re excited to introduce the Open Direct Air Capture 2025 dataset, the largest open dataset for discovering advanced materials that capture CO2 directly from the air. Developed by Meta FAIR, Georgia Tech, and CuspAI, this release enables rapid, accurate screening of carbon

A common takeaway from "the bitter lesson" is we don't need to put effort into encoding inductive biases, we just need compute. Nothing could be further from the truth! Better inductive biases mean better scaling exponents, which means exponential improvements with computation.

Announcing the first workshop on Foundations of Language Model Reasoning (FoRLM) at NeurIPS 2025! 📝Soliciting abstracts that advance foundational understanding of reasoning in language models, from theoretical analyses to rigorous empirical studies. 📆 Deadline: Sept 3, 2025

The best logical explanation of entropy. (It only took me 11 months to find an hour to read it). arxiv.org/abs/2409.09232

I am happy to share the work of our team. The outcome of a collaborative effort, by a joyful group of skilled and determined scientists and engineers! Congrats to the team on this amazing milestone!

The process behind the NeurIPS paper checklist. youtube.com/watch?v=I0KjZk…

Can ancient wisdom promote moral behavior in LLMs? Published with new data! Contemplative insights improve alignment (d = .96) AND boost cooperation + joint reward on the iterated Prisoner’s dilemma task (d = 7+) using GPT-4o + 4.1 Here’s the basic logic 1/x

Looking at the thread. The common frame to look at the more general phenomenon involves an eigenproblem of the form Oƒ = λƒ, where the operator O encodes either: a symmetry (translations, rotations, general group transformations), or a a statistic (e.g. covariance, correlation),

60 years of exponential growth in chip density was achieved not through one breakthrough or technology, but a series of problems solved and new paradigms explored as old ones hit limits. I don't think current AI has hit a wall, but even if it does, there many paths forward now.

x.com/i/article/1961…

The eagle-eyed goat in question being Yuxiang Wei

What if you kept asking an LLM to "make it better"? In some recent work at FAIR, we investigate how we can efficiently use RL to fine-tune LLMs to iteratively self-improve on their previous solutions at inference-time. Training for iterated self-improvement can be costly. The

"Our work is fully reproducible in the sense that it can be reproduced by simulating the laws of physics, starting from the initial configuration of the Universe."

I am proud to announce that today CuspAI has closed a $100M Series A fundraise, co-led by NEA and Temasek, with participation from NVIDIA’s NVentures, Samsung Ventures and @Hyundai_Worldwide.

Excited to share that I am starting a new company dedicated to the creation of verified superintelligence via autoformalization. Building on the amazing RL infrastructure that we have developed at Morph, Math, Inc. has already achieved a breakthrough result by

Last week we found an issue with SWE-Bench, allowing agents to cheat by looking at future commits. Instead of celebrating the SWE-Bench Devs for quickly fixing the issue and being transparent, the HN crowd is dunking on them and drawing wildly inaccurate conclusions about

legendary hacker news comments that aged like milk