Very pleased to share our recent work, in which we use LLMs for automated discovery of interpretable models of animal behavior 🪰🐀🕵️‍♀️ that take the form of Python programs 🐍 See below for a summary of key results by @pcastr.bsky.social! - ThreadSky

Very pleased to share our recent work, in which we use LLMs for automated discovery of interpretable models of animal behavior 🪰🐀🕵️‍♀️ that take the form of Python programs 🐍

See below for a summary of key results by
@pcastr.bsky.social!

Reposted from Pablo Samuel Castro

Can LLMs be used to discover interpretable models of human and animal behavior?🤔

Turns out: yes!

Thrilled to share our latest preprint where we used FunSearch to automatically discover symbolic cognitive models of behavior.
1/12

Comments

hugospiers.bsky.social•15 days ago

Very cool. I need help with some data we have!

lhuntneuro.bsky.social•12 days ago

This is super-fun!! Though I guess even with the rejection sampling approach, we still need DeepMind levels of compute to try it?!?

neurocitizen.bsky.social•12 days ago

Interesting, congrats!

neurokim.bsky.social•15 days ago

This was a fun (and in some ways, mind-bending) project, as we figured out how to make sense of the discovered programs. My favorite part was thinking through these programs, and discussing them with colleagues who've spent a lot of time thinking about RL in psychology + CS.

neurokim.bsky.social•15 days ago

We go through the programs in the paper and point out our favorite snippets (full programs are in the Appendix for the hardcore folks).

neurokim.bsky.social•15 days ago

a HUGE thank you to everyone who contributed -- it really was exhilarating and a privilege to get to work with this group. An extra special thank you to lead
@pcastr.bsky.social and co-senior author
@kevinjmiller.bsky.social, who are very amazing.

neurokim.bsky.social•15 days ago

Tagging the team: @nenadtomasev.bsky.social, Ankit Anand, Navodita Sharma, @neurorishika.bsky.social, Aparna Dev, Kuba Perlin, @siddhantj.bsky.social, @kmlevin.bsky.social, @noemielteto.bsky.social, Will Dabney, Alex Novikov, Glenn Turner, @mariaeckstein.bsky.social, @nathanieldaw.bsky.social

neurokim.bsky.social•15 days ago

Check out the preprint: https://www.biorxiv.org/content/10.1101/2025.02.05.636732v1

kristinmbranson.bsky.social•15 days ago

What do you think about Python as a good, generic form for neurosymbolic AI instead of natural language, as it is human and machine-interpretable, but also exact? We have a UI where we specify the program's behavior in python, and your talk a few months ago got me thinking about this.

Comments

Posting Rules

Reply