1
STOCKTAKE: Measuring the Gap Between Perception and Action in LLM Agents with a Fair Oracle
提出公平oracle量化LLM agent“知道却做不到”的认知鸿沟，揭示感知与行动差距根源，值得关注长期决策评估的研究者深读。
arXiv:2607.13618v1 Announce Type: new Abstract: LLM agents are increasingly evaluated on multi-week decision tasks in which the state that drives cost…