AI的“顿悟”时刻：从玩游戏到省大钱，揭秘智能进化的新路径

本期「人人能懂的AI前沿」，我们重点介绍五篇最新的AI论文：

00:00:27 高手过招：AI是如何在游戏中“悟道”的？

00:04:31 造图的“慢炖”与“快炒”：AI绘画的新思路

00:08:51 AI也懂“看情况办事”了？ 

00:13:34 用对锤子：AI工具的正确使用说明书 

00:18:08 AI点餐的智慧：如何花小钱办大事 

详细论文信息供参考：

[LG] SPIRAL: Self-Play on Zero-Sum Games Incentivizes Reasoning via Multi-Agent Multi-Turn Reinforcement Learning  

[National University of Singapore & A*STAR & Northeastern University]  

https://arxiv.org/abs/2506.24119 

---

[LG] Transition Matching: Scalable and Flexible Generative Modeling  

[Weizmann Institute of Science & FAIR at Meta]  

https://arxiv.org/abs/2506.23589 

---

[LG] Curious Causality-Seeking Agents Learn Meta Causal World  

[Chinese Academy of Sciences & Peking University]  

https://arxiv.org/abs/2506.23068 

---

[LG] Use Sparse Autoencoders to Discover Unknown Concepts, Not to Act on Known Concepts  

[Cornell Tech & UC Berkeley]  

https://arxiv.org/abs/2506.23845 

---

[LG] BEST-Route: Adaptive LLM Routing with Test-Time Optimal Compute  

[The University of British Columbia & Microsoft & Pennsylvania State University]  

https://arxiv.org/abs/2506.22716