00:00:26 高手过招：如何从一堆答案里“攒”出唯一正解？

00:04:59 AI为什么也会“脑补”？

00:10:17 AI刷题的“偏科”陷阱：如何让学霸更有创造力？

00:15:19 一个AI学会做数学题，给了我们什么启发？

00:21:20 目标太多，如何走上“最短路径”？

本期介绍的几篇论文：

[LG] The Majority is not always right: RL training for solution aggregation  

[FAIR at Meta & CMU]  

https://arxiv.org/abs/2509.06870  

---

[LG] From Noise to Narrative: Tracing the Origins of Hallucinations in Transformers  

[Mila - Quebec AI Institute]  

https://arxiv.org/abs/2509.06938  

---

[LG] Outcome-based Exploration for LLM Reasoning  

[FAIR at Meta]  

https://arxiv.org/abs/2509.06941  

---

[LG] Scaling up Multi-Turn Off-Policy RL and Multi-Agent Tree Search for LLM Step-Provers  

[ByteDance Seed]  

https://arxiv.org/abs/2509.06493  

---

[LG] Simple Optimizers for Convex Aligned Multi-Objective Optimization  

[Meta AI & Technion]  

https://arxiv.org/abs/2509.05811