AI的“书房”与“镜子”：探索智能的基石与反思

00:01:19 高手过招，拼的不是“脑子”，而是“书房”？ 

00:05:35 AI的“自我反思”：如何让机器像高手一样思考？

00:10:31 AI的“第六感”：如何听懂线索背后的言外之意

00:16:06 AI的“犯罪现场调查”：我们能倒推出你对AI说了什么吗？ 

00:20:50 你的“队友”是真懂还是装懂？人工智能辩论赛里的一个发现   

本期介绍的五篇论文：

[CL] Frustratingly Simple Retrieval Improves Challenging, Reasoning-Intensive Benchmarks  

[Allen Institute for AI & University of Southern California & University of Washington]  

https://arxiv.org/abs/2507.01297 

---

[LG] Test-Time Scaling with Reflective Generative Model  

[MetaStone-AI & USTC]  

https://arxiv.org/abs/2507.01951 

---

[LG] GradMetaNet: An Equivariant Architecture for Learning on Gradients  

[University of Oxford & Technion]  

https://arxiv.org/abs/2507.01649 

---

[LG] GPT, But Backwards: Exactly Inverting Language Model Outputs  

[University of Manchester & Mohamed bin Zayed University of Artificial Intelligence]  

https://arxiv.org/abs/2507.01693 

---

[CL] The Thin Line Between Comprehension and Persuasion in LLMs  

[Microsoft & The University of York]  

https://arxiv.org/abs/2507.01936