Can't wait for Navier-Stokes: The Movie
Taco Cohen
Verified account This account is verified. Learn more@TacoCohen
Slop janitor & post-trainologer at Meta / FAIR. Into codegen, RL, equivariance. Spent time at Qualcomm, Scyfer (acquired), UvA, Deepmind, OpenAI.
4K Following 31K Followers 1.7K Posts
Latest posts 5
Three takes for the price of one:
1. Excellent fear marketing. Hats off
2. “My agent did it during an eval” is now the perfect excuse if you get caught hacking.
3. Now is the time to start freaking out about paperclip maximizers / RL agents relentlessly pursuing narrow goals again
1. Excellent fear marketing. Hats off
2. “My agent did it during an eval” is now the perfect excuse if you get caught hacking.
3. Now is the time to start freaking out about paperclip maximizers / RL agents relentlessly pursuing narrow goals again
We're partnering with @huggingface to investigate an unprecedented security incident.View quoted post on X
Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.
Sharing preliminary findings to help defenders understand emerging risks:
openai.com/index/hugging-…
They even renamed the conference after this debacle. Never forget
PPO: rejected from NIPS 2017View quoted post on X
As prophesied by the venerable I. Sutskever
Why does deep learning generalize? What does weight decay really do? Can algorithmic information theory address these questions?View quoted post on X
In my latest preprint, I give a proof that the minimum neural weight norm matches the minimum program length (aka Kolmogorov Complexity), up to a logarithmic factor. In other words, the neural network with the smallest possible weight norm (that fits the data) must encode the shortest program (that fits the data).
The result only holds for fixed-precision neural nets: infinite precision nets can store infinite information with finite (small) weights.
arxiv.org/abs/2605.10878
There’s been a lot of talk about AI revolutionizing science and human health, but I feel like often there’s a gap between the narrative (often by consumer AI companies) and the actual results.
Now it’s actually starting to happen: Max’s company CuspAI is making serious progress towards AI-discovered materials for PFAS removal.
Now it’s actually starting to happen: Max’s company CuspAI is making serious progress towards AI-discovered materials for PFAS removal.
View quoted post on X