[人人能懂] 从“一根筋”到“老戏骨”，AI的进化超乎你想象

00:00:31 你的AI有多聪明？关键看它会不会“彩排” 

00:05:06 AI内卷时代，如何找到“天选之子”？ 

00:09:38 AI的“自我修养”：如何让机器自己教自己？ 

00:13:34 给AI装个方向盘，指哪打哪 

00:17:37 AI巨头里的“扫地僧”：没有他，AI秒变“人工智障” 

本期介绍的五篇论文：

[LG] SimuRA: Towards General Goal-Oriented Agent via Simulative Reasoning Architecture with LLM-Based World Model  

[Mohamed bin Zayed University of Artificial Intelligence & Samsung Research]  

arxiv.org  

---

[LG] Consensus-Driven Active Model Selection  

[MIT & UMass Amherst]  

arxiv.org  

---

[CL] CoT-Self-Instruct: Building high-quality synthetic prompts for reasoning and non-reasoning tasks  

[FAIR at Meta & NYU]  

arxiv.org  

---

[CL] Model Directions, Not Words: Mechanistic Topic Models Using Sparse Autoencoders  

[Columbia University]  

arxiv.org  

---

[CL] Unveiling Super Experts in Mixture-of-Experts Large Language Models  

[Meituan & Tsinghua University]  

arxiv.org