作者位次 #3/8ACL 2026Poster· Natural Language Processing
ARES : Adaptive Red-Teaming and End-to-End Repair of Policy-Reward System
Jiacheng Liang · Yao Ma · Tharindu Kumarage · Satyapriya Krishna · Rahul Gupta · Kai-Wei Chang · Aram Galstyan · Charith Peris
摘要
Reinforcement Learning from Human Feedback (RLHF) is central to aligning Large Language Models (LLMs), yet it introduces a critical vulnerability: an imperfect Reward Model (RM) can become a single point of failure when it fails to penalize unsafe behaviors. While existing red-teaming approaches...