RealCompanion：基于长期真实对话推理的人类理解基准
原标题：Paper page - RealCompanion: Benchmarking Human Understanding from Reasoning over Longitudinal Real-World Conversations
AI 摘要
该论文发布了 RealCompanion 基准，包含 10 段真实的人与 AI 伴侣关系、长达 120 天的 27,218 条消息，以及由对话衍生的 profile、persona、chat ground truth 和问题集，每条标注都附带推理链。研究发现：过去信息很少被需要且距离遥远，汇总指标会产生误导；现有检测器无法判断真实消息何时需要记忆，而将消息标注为记忆可将其使用率提高 10 到 14 个百分点；三种 agent 系统以 31 倍成本差异达到相同的 persona 重建 F1。
正文节选
RealCompanion: Benchmarking Human Understanding from Reasoning over Longitudinal Real-World Conversations Abstract A companion that talks with a person for months should come to understand them. It should remember what they said, infer who they are, and know when the past bears on the message in front of it. Testing this requires a real person's record, and such records are private, so benchmarks generate the person and the questions and settle in advance what matters. We release \bench, ten rea