你有没有想过，当AI不再只是一个反应飞快的万事通，而是开始学会“举一反三”，甚至拥有自己的“原则”和“工作流程”时，会发生什么？这一期，我们将看到AI如何自己“开公司”搞科研，又如何建立“中央厨房”模式，用一份力气解决一百个问题。我们还会探讨，如何训练AI坚守原则“不忘初心”，以及它如何模仿人类顶尖专家，像一位真正的科学家那样思考。准备好，让我们一起探寻AI智能正在发生的深刻变革。00:00:38 AI的“一叶知秋”：模型需要读多长的书，才能举一反三？00:05:47 AI的“自我修养”：如何让它学会“不忘初心”？00:10:36 AI的尽头，是开公司？00:14:53 AI的“中央厨房”模式00:20:24 AI当专家，这次可能真不是吹牛本期介绍的几篇论文：[LG] Quantitative Bounds for Length Generalization in Transformers [NEC Labs America & Princeton University & UC Berkeley] https://arxiv.org/abs/2510.27015 ---[LG] Consistency Training Helps Stop Sycophancy and Jailbreaks [Google] https://arxiv.org/abs/2510.27062 ---[LG] The Denario project: Deep knowledge AI agents for scientific discovery [Flatiron Institute & University of Cambridge & Universitat Autonoma de Barcelona] https://arxiv.org/abs/2510.26887 ---[LG] Panprediction: Optimal Predictions for Any Downstream Task and Loss [CMU & UC Berkeley & Columbia University] https://arxiv.org/abs/2510.27638 ---[LG] Glia: A Human-Inspired AI for Automated Systems Design and Optimization [MIT CSAIL] https://arxiv.org/abs/2510.27176

Ask this episode

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

Transcript

Related