本期节目，我们将一起深入AI的大脑，看看它究竟是如何思考的。我们会发现，AI不仅会修炼“内功心法”，以小博大；还会上演复杂的“内心戏”，让我们难辨其思考的真伪。我们还会揭开它通用的“学习公式”，看看聪明的AI为何会掉入“聪明陷阱”，以及它最终如何学会替我们高效地“试错”。00:00:32 AI的“内功心法”：让小模型拥有大智慧的秘密00:05:52 AI的“内心戏”：你看到的思考过程，有多少是“演”的？00:10:49 AI的“聪明陷阱”：为什么懂得多，反而容易犯错？00:16:26 揭秘AI的“学习公式”：原来万变不离其宗00:21:32 让AI替你“试错”，我们能省下多少力气？[CL] Scaling Latent Reasoning via Looped Language Models[ByteDance Seed]https://arxiv.org/abs/2510.25741---[LG] Can Aha Moments Be Fake? Identifying True and Decorative Thinking Steps in Chain-of-Thought[Northeastern University & UC Berkeley]https://arxiv.org/abs/2510.24941---[CL] Are Language Models Efficient Reasoners? A Perspective from Logic Programming[ETH Zürich & EPFL]https://arxiv.org/abs/2510.25626---[CL] Language Model Behavioral Phases are Consistent Across Architecture, Training Data, and Scale[MIT & UCSD]https://arxiv.org/abs/2510.24963---[LG] GPTOpt: Towards Efficient LLM-Based Black-Box Optimization[MIT]https://arxiv.org/abs/2510.25404

Ask this episode

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

Transcript

Related