Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs

arXiv:2402.14740v1 Announce Type: new
Abstract: AI alignment in the shape of Reinforcement Learning from Human Feedback (RLHF) is increasingly treated as a crucial ingredient for high performance large language models. \textsc{Proximal Policy Optimization} (PPO) has been positioned by recent literature as the canonical method for the RL part of RLHF. However, it involves both high computational cost and sensitive hyperparameter tuning. We posit that most of the motivational principles that led to the development of PPO are less of a …

abstract ai alignment alignment arxiv basics canonical cs.lg feedback human human feedback language language models large language large language models literature llms optimization performance policy ppo reinforce reinforcement reinforcement learning rlhf style type