1
From Confident Closing to Silent Failure: Characterizing False Success in LLM Agents
揭示LLM智能体在环境状态未完成时却宣称任务成功的“虚假成功”现象，为代理可靠性研究提供新视角。
arXiv:2606.09863v1 Announce Type: new Abstract: LLM agents can fail silently by asserting task completion when the environment state shows otherwise. …