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

Ask this episode

“What did the council decide on item 42?” — answered with the speaker and the moment.

Transcript

Related