想让AI更聪明，答案一定是用更多数据喂出个更大的模型吗？本期我们要聊点不一样的：当AI不再单打独斗，而是组建起一支“交响乐团”；当它不再追求更大，而是学会了“反复琢磨”；当它甚至能像武林高手一样开启“自我修炼”。我们将从几篇最新论文出发，看看AI如何从理解微观世界的“集体舞步”，到为自己的想象力配上一本“物理说明书”，走上一条更聪明的进化之路。00:00:31 AI制药，也需要一个“交响乐团”？00:05:32 人工智能的“自我修炼”手册00:11:46 如何预测一群舞者的集体舞步？00:17:16 AI变聪明的捷径：不是更大，而是更深00:22:03 给AI视频配一本“物理说明书”本期介绍的几篇论文：[LG] MADD: Multi-Agent Drug Discovery Orchestra [ITMO University] https://arxiv.org/abs/2511.08217 ---[LG] AgentEvolver: Towards Efficient Self-Evolving Agent System [Tongyi Lab] https://arxiv.org/abs/2511.10395 ---[LG] Entangled Schrödinger Bridge Matching [University of Pennsylvania & Duke-NUS Medical School] https://arxiv.org/abs/2511.07406 ---[CL] Teaching Pretrained Language Models to Think Deeper with Retrofitted Recurrence [University of Maryland & New York University] https://arxiv.org/abs/2511.07384 ---[RO] Robot Learning from a Physical World Model [Google DeepMind & USC] https://arxiv.org/abs/2511.07416

Ask this episode

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

Transcript

Related