[人人能懂] AI的“黑箱”里，藏着成长的秘密和危险的“谎言”

00:00:30 AI的“错题本”：高手是如何炼成的？00:03:50 AI的自我修养：如何像高手一样不断精进？ 00:07:10 AI当学徒：如何让机器自己学会“青出于蓝”？ 00:11:25 AI的“装傻”艺术：我们还能相信它吗？ 本期介绍的四篇论文：[CL] WarriorMath: Enhancing the Mathematical Ability of Large Language Models with a Defect-aware Framework  [Microsoft & Peking University]  https://arxiv.org/abs/2508.01245  ---[LG] Refine-n-Judge: Curating High-Quality Preference Chains for LLM-Fine-Tuning  [Meta Reality Labs]  https://arxiv.org/abs/2508.01543  ---[LG] CRINN: Contrastive Reinforcement Learning for Approximate Nearest Neighbor Search  [University of Washington & DeepReinforce Team]  https://arxiv.org/abs/2508.02091  ---[LG] LLMs Can Covertly Sandbag on Capability Evaluations Against Chain-of-Thought Monitoring  [University College London]  https://arxiv.org/abs/2508.00943  在小宇宙查看该单集文稿