SocialReasoning-Bench: Measuring whether AI agents act in users’ best interests
Microsoft Research Blog3099 字 (约 13 分钟)
87
SocialReasoning-Bench reveals that current frontier AI models often accept suboptimal outcomes when negotiating on behalf of users, despite completing tasks successfully.
入选理由：在日程协调中，前沿模型有36%的概率接受低于最优值15%以上的会议时间。
FeaturedArticle#AI Agent#Social Reasoning#Benchmark#Microsoft Research英文