今天我们来聊聊，为什么AI的创意越来越像“标准答案”，仿佛陷入了一个巨大的“蜂巢思维”？我们将探讨如何跳出这个怪圈，让AI学会使用计算器进行事实核查，甚至培养出程序员般的代码“审美”。更进一步，我们会揭示AI学习的两种反常识秘诀：一是故意让它犯“高质量”的错误，二是用“分而治之”的策略去攻克马拉松式的超长任务。从思维定式到品味养成，再到学习心法，本期将带你看到AI如何变得更“聪明”，也更“人性化”。00:00:39 人工智能的“标准答案”陷阱00:05:54 让AI裁判学会用“计算器”00:10:37 想让AI更聪明？先教它犯“高质量”的错00:15:05 让AI学会“好看”：一个程序员的品味是如何炼成的00:19:42 想跑赢马拉松？别从第一步开始练本期介绍的几篇论文：[CL] Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond)[University of Washington]https://arxiv.org/abs/2510.22954---[CL] Incentivizing Agentic Reasoning in LLM Judges via Tool-Integrated Reinforcement Learning[Google Cloud AI Research]https://arxiv.org/abs/2510.23038---[LG] BugPilot: Complex Bug Generation for Efficient Learning of SWE Skills[Cornell University & University of California San Diego & University of North Carolina at Chapel Hill]https://arxiv.org/abs/2510.19898---[CL] Code Aesthetics with Agentic Reward Feedback[Microsoft Research Asia]https://arxiv.org/abs/2510.23272---[LG] Transitive RL: Value Learning via Divide and Conquer[UC Berkeley]https://arxiv.org/abs/2510.22512

Ask this episode

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

Transcript

Related