一作ACL 2026Poster· Natural Language Processing
No More Stale Feedback: Co-Evolving Critics for Open-World Agent Learning
Zhicong Li · Lingjie Jiang · Yulan Hu · Xingchen Zeng · Yixia Li · Xiangwen Zhang · Guanhua Chen · Zheng Pan · Xin Li · Yong Liu
摘要
Critique-guided reinforcement learning (RL) has emerged as a powerful paradigm for training LLM agents by augmenting sparse outcome rewards with natural-language feedback. However, current methods often rely on static or offline critic models, which fail to adapt as the policy evolves. In on-poli...