AI RESEARCH
Curriculum-RLAIF: Curriculum Alignment with Reinforcement Learning from AI Feedback
arXiv CS.AI
•
ArXi:2505.20075v2 Announce Type: replace 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 mismatch of overly challenging samples with model capacity.