Tag
论文提出 PersonaCUABench，通过 74 个任务和 44 个不同用户画像来评估计算机使用代理（CUA），发现相比原始指令，任务完成率平均下降 36%，表明 CUA 在满足真实用户的个性化需求和约束方面仍存在显著不足。
A user shares their positive experience with Anthropic's Opus 5.5 AI model, stating it has become their daily driver and expressing happiness about its release.
The article criticizes Altman's controversial actions regarding shutting down applications and handling refunds and tokens, leading to user reluctance and dissatisfaction.
The paper introduces MemUse, a benchmark that shows direct QA accuracy for conversational memory does not predict user satisfaction, whereas natural integration of prior context does, revealing a large gap between recall and conversational use.
MapSatisfyBench is a benchmark for evaluating LLM-based map agents on their ability to recover implicit decision factors from underspecified user queries, shifting evaluation from task completion to satisfaction-aware spatial decision making.