本期我们要聊的几篇最新论文，简直就像是AI上演了一出精彩的“进化三重奏”。你有没有想过，AI不仅能亲自下场做实验，还能通过扔掉海量信息反而学得更快？我们还会看到，AI如何像一个不眠不休的科研团队那样在失败中进化，像军队一样高效分工，以及这一切的背后，如何靠一本“备忘录”将所有经验沉淀为真正的智慧。准备好了吗？让我们一起探索AI正在解锁的全新可能性！

00:00:34 AI 不再只是“思想家”，它开始“动手”了

00:04:59 AI 进化新思路，扔掉 95% 的信息，反而学得更好？

00:10:55 AI的“试错”进化论

00:17:42 将军与士兵，人工智能的完美分工

00:22:55 给AI装个“备忘录”，为什么笨办法反而是真聪明？

本期介绍的几篇论文：

[AI] Accelerating Scientific Research with Gemini in the Real-World

[Google DeepMind & Duke University & Columbia University]

https://arxiv.org/abs/2608.26701

---

[CV] LeVJEPA: Efficient & Scalable Video Pretraining without the Heuristics

[German Cancer Research Center & Mila]

https://arxiv.org/abs/2608.27395

---

[AI] AgentFold: Closed-Loop Agentic Search for Protein Folding Model Design

[Hunan University & Nanjing University & The Chinese University of Hong Kong]

https://arxiv.org/abs/2608.26747

---

[AI] Decoupling Planning and Control for Instructable Agents

[UC Berkeley & Google DeepMind]

https://arxiv.org/abs/2608.26788

---

[AI] WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolution

[Google Research]

https://arxiv.org/abs/2608.27454