[人人能懂] AI私房话：它的顿悟、纠结与学习心法

00:00:34 AI也会“想多了”？聊聊机器的“顿悟”与“纠结”

00:05:23 AI的“顿悟”时刻：为什么高手总在“瓶颈期”后迎来爆发？

00:10:29 AI的大脑换食谱：不吃“山珍海味”，改吃“压缩饼干”

00:16:16 给复杂世界建模：AI如何成为科学家的“超级学徒”

00:20:44 顶尖高手的心法：如何一眼看出谁是“好学生”？

本期介绍的五篇论文：

[LG] MDPO: Overcoming the Training-Inference Divide of Masked Diffusion Language Models  

[University of Tübingen]  

https://arxiv.org/abs/2508.13148  

---

[LG] Learning In-context n-grams with Transformers: Sub-n-grams Are Near-stationary Points  

[EPFL]  

https://arxiv.org/abs/2508.12837  

---

[CV] Separating Knowledge and Perception with Procedural Data  

[MIT]  

https://arxiv.org/abs/2508.11697  

---

[LG] Simulation-Based Inference: A Practical Guide  

[University of Tübingen]  

https://arxiv.org/abs/2508.12939  

---

[CV] Dextr: Zero-Shot Neural Architecture Search with Singular Value Decomposition and Extrinsic Curvature  

[Friedrich-Alexander-Universität Erlangen-Nürnberg & Universität Ulm]  

https://arxiv.org/abs/2508.12977