一作ACL 2026Findings· Findings of ACL
CAR : Empowering Agents with Dynamic Tool Synthesis and Global Trajectory Rectification
Kai Wang · Xu Wang · Zhiyuan Fu · Yudong Zhang · Yang Wang
摘要
While current LLM agents utilizing paradigms like ReAct or Plan-and-Solve have established a strong foundation for step-by-step reasoning, they remain brittle in open-ended environments due to two intrinsic limitations: (1) A closed action space: These frameworks are confined to static, pre-defin...