今天我们要聊的话题，比你想象的更微妙：如何与一个既强大又有点“怪脾气”的AI共事？本期节目，我们将从几篇最新论文出发，看看如何不打开“黑箱”就把机器人训练成顶尖高手；为何让AI“三思而后行”反而可能把事情搞砸；以及如何像一位高明的项目经理，管好那个才华横溢却总爱“自由发挥”的AI程序员。准备好了吗？让我们一起探索驾驭AI的全新智慧。

00:00:32 不开箱，如何把一个通用机器人，训练成顶尖高手？

00:06:45 让AI“三思而后行”，为什么结果可能更糟？

00:13:22 想让AI学得好，教它“目标”还是教它“动作”？

00:19:49 AI在思考时，到底有多“用力”？

00:24:50 AI队友，如何管好一个“不听话”的天才

本期介绍的几篇论文：

[RO] CLIFT: Turning Gemini Robotics On-Device into Humanoid Specialists via Non-Invasive Closed-Loop Iterative Fine-Tuning

[UC Berkeley]

https://arxiv.org/abs/2607.29172

---

[LG] Reflection or Re-Generation? Why LLM Revision Fails Where Human Revision Succeeds

[Amazon]

https://arxiv.org/abs/2607.28908

---

[LG] When Does On-Policy Interaction Help? Representational Tradeoffs in Value-Based Imitation Learning

[EPFL]

https://arxiv.org/abs/2607.29617

---

[AI] How Hard Does It Think? Analyzing Step-Aware Reasoning Energy in LLM Chain-of-Thought Trajectories

[UC Merced & UC San Diego]

https://arxiv.org/abs/2607.28674

---

[AI] From Code Review to Code Critique: Intent, Drift, and Spotlight for AI-Generated Diffs at Scale

[Meta & Concordia University]

https://arxiv.org/abs/2607.29516