[人人能懂] AI思考的五种新姿势：它如何学会了“死磕”、“变通”与“偷懒”？

00:00:34 AI侦探养成记：如何让机器学会“死磕到底”？00:04:22 AI也需要“元学习”：如何打造一把能开万能锁的钥匙？00:07:56 拆解AI大脑：它如何学会“绕个弯”解决问题？00:12:21 AI学会“举一反三”的秘密：两层楼就够了？00:16:34 AI思考的秘密：为什么“少”就是“多”？本期介绍的五篇论文：[CL] Beyond Ten Turns: Unlocking Long-Horizon Agentic Search with Large-Scale Asynchronous RL  [Tsinghua University]  https://arxiv.org/abs/2508.079  ---[LG] AdaptFlow: Adaptive Workflow Optimization via Meta-Learning  [Peking University & University of Chinese Academy of Sciences]  https://arxiv.org/abs/2508.08053  ---[LG] Multi-head Transformers Provably Learn Symbolic Multi-step Reasoning via Gradient Descent  [CMU & UPenn & OSU]  https://arxiv.org/abs/2508.08222  ---[LG] What One Cannot, Two Can: Two-Layer Transformers Provably Represent Induction Heads on Any-Order Markov Chains  [MIT & EPFL & UC Berkeley]  https://arxiv.org/abs/2508.07208  ---[CL] Less Is More: Training-Free Sparse Attention with Global Locality for Efficient Reasoning  [Princeton University & CMU]  https://arxiv.org/abs/2508.07101  在小宇宙查看该单集文稿