Center for Human-Compatible AI

Highway driving simulation in JAX for Reinforcement Learning research

Dataset for the paper: Political Neutrality as Balanced Approval: A Large-Scale Human Evaluation of AI Responses

Testing ranking algorithms to improve social cohesion

A benchmark environment for fully cooperative human-AI performance.

A prompt injection game to collect data for robust ML research

Clean PyTorch implementations of imitation and reward learning algorithms

Prosocial Ranking Challenge Perspective Ranker

Code for "Evidence of Learned Look-Ahead in a Chess-Playing Neural Network"