Creston Brooks
139 posts

Creston Brooks
@crestonbrooks
building ~ previously @SentientAGI @princeton_nlp




that’s it. i’m only hiring classicists now.





My own framing on this: we are post-Turing Test but pre-AGI. Possibly the weirdest 5-10 year period humanity will ever face.





In an AI world, effort is no longer the bottleneck—ideas are. Open-source AI enables anyone to have access to build anything they want without depending on the one-sided gates that closed-source offers. This is amplified by the open innovation that open-source development enables.

Most RL frameworks are fundamentally unstable. We wasted more H100 hours on debugging this than any other issue fornour multi-turn, multi-env RL run (below). When using OpenAI-style messages for env interactions, parsing and retokenizing leads to subtly different tokens. This creates extremely unlikely tokens, which dominate the gradient and over time lead to collapse. The screenshots describe the mechanism in more detail. We tried a lot of interventions, but ended up reimplementing our environments to use token lists directly (Tokens-in/Tokens-out). This fixed it immediately. Always inspect logprobs!






