AI agents are usually evaluated by whether they complete a task. In interactive service settings, a successful agent can still frustrate users by asking repe…
机构:滴滴
来源:arXiv 2609.17985 | AI4Papers 论文推荐平台