LLM Active Alignment: A Nash Equilibrium Perspective

Tonghan Wang, Yuqi Pan, Xinyi Yang +3

We develop a game-theoretic framework for predicting and steering the behavior of populations of large language models (LLMs) through Nash equilibrium (NE) analysis. To avoid the i…