今天我们不只聊AI能做什么，而是要深入它的“内心”，看看它是如何思考和学习的。我们将一起探索，AI如何在信息不足的“战争迷雾”中做出超人决策，又如何像滚雪球一样自我进化，解决超长难题。我们还会看到，AI如何被“逼”着建立起真正的内心世界，甚至它的训练过程，竟然能用中学物理的理想气体定律来解释！准备好了吗？让我们一起揭开这些聪明策略的神秘面纱，看看AI如何实现从“死记硬背”到“融会贯通”，从“慢工细活”到“快马绣花”的华丽变身。00:00:39 信息不足，如何做出“超人”决策？00:06:35 AI如何学会“举一反三”？00:12:54 AI的“顿悟”：如何让机器不只记忆，更懂世界00:17:08 AI炼丹炉里的理想气体00:22:22 AI训练的“既要又要”：如何让快马也能绣花？本期介绍的几篇论文：[LG] Superhuman AI for Stratego Using Self-Play Reinforcement Learning and Test-Time Search[CMU & NYU Tandon School of Engineering & Stanford University]https://arxiv.org/abs/2511.07312---[LG] Transformers Provably Learn Chain-of-Thought Reasoning with Length Generalization[University of Pennsylvania & CMU]https://arxiv.org/abs/2511.07378---[LG] Next-Latent Prediction Transformers Learn Compact World Models[Microsoft Research]https://arxiv.org/abs/2511.05963---[LG] Can Training Dynamics of Scale-Invariant Neural Networks Be Explained by the Thermodynamics of an Ideal Gas?[Constructor University & Mila]https://arxiv.org/abs/2511.07308---[LG] TNT: Improving Chunkwise Training for Test-Time Memorization[Google Research & University of Southern California]https://arxiv.org/abs/2511.07343

Ask this episode

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

Transcript

Related