你有没有想过，两个完全不同的AI高手，不用原始数据就能“合体”成全能大侠？甚至还能拥有一本超级“错题本”，学会真正的举一反三？更进一步，当AI员工比我们还聪明时，我们又该如何设计一套“管理学”，用一个聪明的“路由器”来调度一群天才？但在这看似强大的智慧背后，AI究竟是像好老师一样“因材施教”，还是在巧妙地“装懂”？本期节目，我们将通过五篇最新的AI论文，一起探索AI的群体智慧与心智幻觉。00:00:38 AI模型也能合体？不用数据，照样让你武功大增00:04:53 让AI学会举一反三，需要一本怎样的“错题本”？00:10:12 AI界的“诸葛亮”，如何给你“三个臭皮匠”的智慧？00:14:41 AI当老师，到底是真懂你，还是在“装懂”？00:21:40 当你的员工比你还聪明，该怎么管？本期介绍的几篇论文：[LG] Model Merging via Data-Free Covariance Estimation[Universite de Montr ́eal & University of Toronto]https://arxiv.org/abs/2604.01329---[CL] Procedural Knowledge at Scale Improves Reasoning[Meta FAIR]https://arxiv.org/abs/2604.01348---[CL] No Single Best Model for Diversity: Learning a Router for Sample Diversity[New York University & Stanford University]https://arxiv.org/abs/2604.02319---[AI] Do Large Language Models Mentalize When They Teach?[Princeton University & New York University]https://arxiv.org/abs/2604.01594---[LG] CORAL: Towards Autonomous Multi-Agent Evolution for Open-Ended Discovery[MIT & NUS]https://arxiv.org/abs/2604.01658

Ask this episode

“What did the council decide on item 42?” — answered with the speaker and the moment.

Transcript

Related