Research update: Still getting settled

4 views

Skip to first unread message

Ram Rachum

unread,

Oct 1, 2025, 6:07:52 AM (3 days ago) Oct 1

to ram-rachum-res...@googlegroups.com

Hi everyone!

It's been almost a month since I got back to Israel. It took me a while to get used to my old life. This will be a short update.

Breakdown and steerage

Last month I told you that I'm pivoting from my work on social behavior of agents to an explainability project. I developed a technique that assists in answering the question "why did my RL agent learn to behave in such a way?" For any behavior metric that you define, it's able to find the timesteps that an RL agent trained on that encouraged it the most to learn that specific behavior. I'll have more details on this project as it progresses.

A Conservative Vision For AI Alignment

Last month I sent you the first post in the LessWrong sequence that David Manheim and I are writing. I hoped that the second post would already be ready by the time I send this update, but we're still working on it. I believe that by next month's update both the second and third posts will be published, and I'll link to them.

Freelance work

It's been a while since I made money or had any funding for my research. I'm now starting to do part-time freelance work on LLM evaluations. I'm happy about this because it's a chance to fund my research while doing work that is somewhat close to my research interests. It does mean I'll have less time to work on the actual research in the next two months, but it also means I won't starve, which is a plus.

Also, I'm applying to a few different internship programs, both for the winter and summer.