Self-Play Preference Optimization for Language Model Alignment

arXiv:2405.00675v1 Announce Type: new
Abstract: Traditional reinforcement learning from human feedback (RLHF) approaches relying on parametric models like the Bradley-Terry model fall short in capturing the intransitivity and irrationality in human preferences. Recent advancements suggest that directly working with preference probabilities can yield a more accurate reflection of human preferences, enabling more flexible and accurate language model alignment. In this paper, we propose a self-play-based method for language model alignment, which treats the problem as a constant-sum two-player game aimed …

abstract alignment arxiv cs.ai cs.cl cs.lg enabling feedback human human feedback language language model optimization parametric reinforcement reinforcement learning rlhf self-play stat.ml type