今天我们要聊一个根本问题：聪明的AI到底是什么样？是像一位能极速压缩思考时间的解题大师，还是一个能在想象的沙盒里自我进化的机器人？我们还会一起探究，为什么AI时而是无所不知的学霸，时而又是死记硬背的书呆子，甚至还会一本正经地胡说八道。最后，你会发现，让AI变得更懂你的终极秘诀，可能恰恰是先教会它如何“装傻”提问。让我们跟随几篇最新论文，一起解剖AI的思考内核。00:00:36 AI变聪明的秘密：不是知道更多，而是想得更快00:06:29 给机器人一个“沙盒”，让它在想象中进化00:12:37 为什么AI既是学霸，又是书呆子？00:17:19 AI的“发挥失常”：一个问题，两种症状00:23:35 如何让机器更懂你？答案是：先让它学会“装傻”本期介绍的几篇论文：[LG] AI Agents as Universal Task Solvers: It’s All About Time[AWS Agentic AI]https://arxiv.org/abs/2510.12066---[RO] Ctrl-World: A Controllable Generative World Model for Robot Manipulation[Stanford University & Tsinghua University]https://arxiv.org/abs/2510.10125---[LG] LLM Knowledge is Brittle: Truthfulness Representations Rely on Superficial Resemblance[FAIR at Meta & University of Zurich]https://arxiv.org/abs/2510.11905---[CL] Generation Space Size: Understanding and Calibrating Open-Endedness of LLM Generations[Stanford University]https://arxiv.org/abs/2510.12699---[LG] Asking Clarifying Questions for Preference Elicitation With Large Language Models[Google]https://arxiv.org/abs/2510.12015

Ask this episode

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

Transcript

Related