Toward Optimal LLM Alignments Using Two-Player Games

Rui Zheng, Hongyi Guo, Zhihan Liu +10

The standard Reinforcement Learning from Human Feedback (RLHF) framework primarily focuses on optimizing the performance of large language models using pre-collected prompts. Howev…