作者位次 #4/7ACL 2026Findings· Findings of ACL
Curriculum- RLAIF : Curriculum Alignment with Reinforcement Learning from AI Feedback
Jiaye Lin · Mengdi Li · Xufeng Zhao · Wenhao Lu · Peilin Zhao · Stefan Wermter · Di Wang
摘要
Reward models trained through Reinforcement Learning from AI Feedback (RLAIF) methods frequently suffer from limited generalizability, which hinders the alignment performance of policy models. This challenge stems from various issues, including distribution shift, preference label noise, and mism...