We just released a preprint on Discovering Symbolic Cognitive Models from Human and Animal Behavior where we explore these ideas in the field of neuroscience:

www.biorxiv.org/content/10.1...

In chess, our planning agents effectively reach grandmaster-level strength with a comparable search budget to that of human players, achieving strong play against existing bots, and in the paper we share a number of fascinating games that our agents played in both Chess and Chess960, in the appendix

The underlying language model is trained to capture the transition and value functions across games.

In external search, the model guides Monte Carlo Tree Search (MCTS) rollouts and evaluations without calls to an external engine, and in internal search, the model directly generates in-context a linearized tree of potential futures and a resulting final choice.

Super excited to be here for the world chess championship - and if game one is anything to go by, it's going to be quite an exciting match!