Uncovering Robustness Weaknesses in LLM Agents via Persona Evolution with Task-Integrity Constraints
Interactive-agent evaluations commonly rely on standardized LLM-based user simulators that exhibit broadly cooperative behavior, potentially overlooking failures that arise under diverse but realistic user interactions. This can lead to overly optimistic estimates of agent robustness and provide limited failure signals...
Yu-Tao Xie, Jing-Bo Yang, Guanyu Yao et al.
· 0 citations