00:00:36 AI的“回音室”危机：当复印件开始复印自己00:05:30 AI的“超级大脑”里，谁在干活，谁在摸鱼？00:10:23 AI的“隐形纹身”：一张纸巾就能擦掉？00:15:27 AI的“选择之道”：为何“条条大路通罗马”本身就是一种智慧？00:20:39 AI的“开窍”秘诀：好学生为何不能太“自信”？本期介绍的几篇论文：[LG] ForTIFAI: Fending Off Recursive Training Induced Failure for AI Models [UC San Diego & Stanford University] https://arxiv.org/abs/2509.08972 ---[CL] Steering MoE LLMs via Expert (De)Activation [University of California, Los Angeles & LMU Munich & Adobe Research] https://arxiv.org/abs/2509.09660 ---[LG] Character-Level Perturbations Disrupt LLM Watermarks [University of Technology Sydney & Griffith University] https://arxiv.org/abs/2509.09112 ---[LG] An entropy formula for the Deep Linear Network [G Menon, T Yu] https://arxiv.org/abs/2509.09088 ---[CL] CDE: Curiosity-Driven Exploration for Efficient Reinforcement Learning in Large Language Models [Tencent AI Lab] https://arxiv.org/abs/2509.09675

Ask this episode

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

Transcript

Related