如果AI不再靠“题海战术”，用小教练就能撬动超级大脑，甚至1024个AI在没有老板的情况下能自发组织协作，世界会变成怎样？本期节目，我们将深入最新论文，带你围观AI何时在靠写步骤“装模作样”、何时又是真正的“思维承重”。我们还会看到懂编程的Agent如何用上帝视角的统计反思秒杀盲目试错，以及科学家怎样用精妙的“套娃归因”在千亿神经元里一秒揪出掌权者。5篇最新论文，带你穿透技术黑盒，看清智能进化照见的人类认知与协作智慧！00:00:40 别让聪明的大脑陷入“题海战术”，一次关于AI重塑思考方式的启示00:05:25 放弃“超级大脑”的执念，当1024个AI决定自己管理自己，真正的启发来了00:10:19 AI写的“解题步骤”，到底是真思考还是在做戏？00:16:42 别再盲目试错了，跳出局部陷阱的“上帝视角”工作法00:20:53 如何在一个极其复杂的系统里，精准揪出那个“说了算”的人？本期介绍的几篇论文：[CL] Beyond Repeated Sampling: Learning Search Policies for LLM Reasoning[Meta FAIR & Université Paris-Sacla]https://arxiv.org/abs/2609.26704 ---[CL] Agensh: Scaling Organizational Intelligence to 1,024 Agents[Microsoft Research]https://arxiv.org/abs/2609.26781 ---[AI] From Decorative to Load-Bearing: Task Difficulty Shapes the Causal Role of Chain-of-Thought[Cornell University & CMU]https://arxiv.org/abs/2609.25366 ---[AI] Coding Agents are Strong Prompt Optimizers[Microsoft]https://arxiv.org/abs/2609.26261 ---[CL] Matryoshka attribution: Learning to attribute language model outputs to representations and weights[Stanford University]https://arxiv.org/abs/2609.25518 在小宇宙查看该单集文稿

Ask this episode

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

Transcript

Related