HAI & SDS Seminar with with Wolfgang Lehrach

Code World Models for General Game Playing

Visit our website to learn more about the event agenda, speakers, and other details

While Large Language Models (LLMs) show promise in many domains, relying on them for direct policy generation in games often results in illegal moves and poor strategic play. In this talk, I present an approach that moves away from direct prompting, instead using LLMs as program synthesizers to bridge the gap between natural language rules and symbolic world models.  The LLM receives a game description and example trajectories, and outputs an executable, symbolic world model (CWM) represented in Python. The trajectories also ensure the rules are correctly captured and aid in refining the CWM if they are not.  Note that even trajectories containing only a single player's observations and actions can be used to help validate and refine CWMs.  Furthermore, partially observed trajectories also allow comparisons between CWMs via a bound on the likelihood.  

Given a CWM, Monte Carlo Tree Search (MCTS) or Reinforcement Learning (RL) methods can play the game, and gameplay can be further enhanced by adding in LLM-derived synthesized value functions. Imperfect information games are handled by having the LLM synthesize inference functions to impute information sets, or by directly training reinforcement learning policies on top of the CWM.

Details:

Time: 12:00 pm - 1:15 pm PT

Location: Gates Computer Science Building, Room 119, 353 Jane Stanford Way, CA 94503.

Advance registration is required to guarantee entry.

We are happy to accommodate walk-in guests as space allows. Please note that if the seminar is sold out, we may not be able to admit those without prior registration.

Know what’s Happening Next — before everyone else does.

Get App

Host or Publisher Stanford Institute for Human-Centered Artificial Intelligence (HAI)