你有没有想过，AI写文章不靠“串珠子”，而是像雕塑家一样从混沌中“凿”出杰作？为什么把两个最先进的技术组合，反而会得到一个最差的结果？本期节目，我们将一起探讨几篇最新论文，看看AI如何像一个聪明的团队，在网上冲浪却不“学坏”；如何在与人类“上有政策，下有对策”的博弈中保持学习能力；以及它如何像一个决策守门员，把宝贵的精力只花在最值得的探索上。00:00:32 AI写作新思路，像雕塑家一样“凿”出好文章00:05:19 AI的“阅读”困境，该读字，还是读字母？00:11:41 你的“最强大脑”，差的不是智商，而是搜索方法00:17:05 AI时代的“上有政策，下有对策”00:22:37 决策的“守门员”，如何把精力花在刀刃上本期介绍的几篇论文：[CL] ELF: Embedded Language Flows [MIT] https://arxiv.org/abs/2605.10938 ---[LG] The Efficiency Gap in Byte Modeling [Google DeepMind] https://arxiv.org/abs/2605.12928 ---[CL] Context Training with Active Information Seeking [Google DeepMind] https://arxiv.org/abs/2605.13050 ---[LG] Strategic PAC Learnability via Geometric Definability [Technion – Israel Institute of Technology] https://arxiv.org/abs/2605.13426 ---[LG] Delightful Exploration [Google DeepMind] https://arxiv.org/abs/2605.13287

Ask this episode

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

Transcript

Related