00:00:32 你的夸奖，正在“毒害”AI 

00:05:22 数据大扫除：不止是扔垃圾，更是换风格

00:10:55 AI的“世界观”：它如何从零开始看懂现实？ 

00:15:46 AI的“省钱攻略”：如何花小钱办大事？ 

00:20:27 喂养AI的新艺术：从“吃什么”到“怎么吃” 

本期介绍的无篇文章：

[LG] Off-Policy Corrected Reward Modeling for Reinforcement Learning from Human Feedback  

[The University of Tokyo and RIKEN AIP]  

https://arxiv.org/abs/2507.15507  

---

[LG] Distributional Unlearning: Forgetting Distributions, Not Just Samples  

[EPFL & Stanford University]  

https://arxiv.org/abs/2507.15112  

---

[LG] Skill Learning via Policy Diversity Yields Identifiable Representations for Reinforcement Learning  

[Max Planck Institute for Intelligent Systems & University of Tübingen]  

https://arxiv.org/abs/2507.14748  

---

[CL] Towards Compute-Optimal Many-Shot In-Context Learning  

[Google Cloud AI Research]  

https://arxiv.org/abs/2507.16217  

---

[LG] LLM Data Selection and Utilization via Dynamic Bi-level Optimization  

[University of Chinese Academy of Sciences & Huawei Noah’s Ark Lab]  

https://arxiv.org/abs/2507.16178