Contact

Email is the best way to reach me. If you are writing about a paper or the code, say which one.

What to reach out about

• Interpretability collaborations, especially on reward models or on what finetuning does to a model
• Mistakes in a paper, a post, or the code
• A paper or a direction I should be reading
• Research roles and AI safety work

I typically respond within 2 to 3 days. If you do not hear back within a week, send a follow-up; sometimes messages get lost.