[人人能懂] 驯兽师、陪练与裁判：我们如何塑造更聪明的AI？

00:00:35 我们永远无法根除AI的“幻觉”，但可以学会与它共舞 00:04:28 人工智能的“笨功夫”：一个鸟类识别模型教给我们的事00:08:44 AI世界的“计分板”，正在悄悄升级00:12:25 AI如何学会当数学家？三个你也能用的“笨”办法00:17:15 AI码农进化论：如何“调教”一个更聪明的程序员？本期介绍的五篇论文：[CL] A comprehensive taxonomy of hallucinations in Large Language Models  [Universitat de Barcelona]  https://arxiv.org/abs/2508.01781  ---[LG] Perch 2.0: The Bittern Lesson for Bioacoustics  [Google DeepMind]  https://arxiv.org/abs/2508.04665  ---[CL] CompassVerifier: A Unified and Robust Verifier for LLMs Evaluation and Outcome Reward  [Shanghai AI Laboratory]  https://arxiv.org/abs/2508.03686  ---[LG] Goedel-Prover-V2: Scaling Formal Theorem Proving with Scaffolded Data Synthesis and Self-Correction  [Princeton University & Tsinghua University]  https://arxiv.org/abs/2508.03613  ---[LG] Training Long-Context, Multi-Turn Software Engineering Agents with Reinforcement Learning  [Nebius AI]  https://arxiv.org/abs/2508.03501  在小宇宙查看该单集文稿