你有没有想过，未来的AI不仅能回答你的问题，还能从与你的每一次互动中汲取经验，悄悄进化？它甚至还能在犯错后“自我反思”，像我们一样“长记性”。本期我们将一起探索几篇最新论文，看看AI如何学会像一个聪明的“CEO”一样管理自己的思考，如何通过精准“剪枝”在你的手机里狂飙，以及如何消灭那些你看不到的“计算成本”，变得更高效、更智慧。

00:00:32 AI进化论，为什么你的“差评”正在喂养一个更聪明的它

00:05:19 让AI在手机里狂飙，快，才是一切

00:10:38 AI提速19%的秘密，你以为的计算，其实是搬运

00:15:20 AI犯了错，能不能让它自己“长记性”？

00:21:26 你的大脑里，缺一个聪明的“CEO”

本期介绍的几篇论文：

[CL] Online Experiential Learning for Language Models  

[Microsoft Research]  

https://arxiv.org/abs/2603.16856

---

[LG] MobileLLM-Flash: Latency-Guided On-Device LLM Design for Industry Scale  

[Meta AI]  

https://arxiv.org/abs/2603.15954

---

[LG] FlashSampling: Fast and Memory-Efficient Exact Sampling  

[LMU Munich & FlashSampling & Princeton University]  

https://arxiv.org/abs/2603.15854

---

[LG] Meta-TTRL: A Metacognitive Framework for Self-Improving Test-Time Reinforcement Learning in Unified Multimodal Models  

[Tsinghua University & JD.COM]  

https://arxiv.org/abs/2603.15724

---

[RO] When Should a Robot Think? Resource-Aware Reasoning via Reinforcement Learning for Embodied Robotic Decision-Making  

[CMU & Northeastern University & Harvard University]  

https://arxiv.org/abs/2603.16673