今天，我们来聊一次AI的“认知升级”，它已经不满足于简单地听从指令了。当AI开始自己“进化”出新算法，我们该如何绘制它创造的知识地图？当AI的考试不再是答题，而是“活下去”，我们又该如何成为一名能随时修正航向的“舵手”，甚至看懂它藏在心中的“锦囊妙计”？本期节目，就让我们通过几篇最新论文，一窥AI智能的未来形态。00:00:32 AI进化论：让算法自己发现算法00:05:38 科学研究的GPS：如何看透一个陌生领域？00:11:13 AI 的下一场考试，考的是「活下去」的能力00:16:22 别让AI瞎跑，你得学会当个好舵手00:20:52 给AI一个“锦囊”，它就能变得更聪明？本期介绍的几篇论文：[LG] Discovering state-of-the-art reinforcement learning algorithms [Google DeepMind] https://www.nature.com/articles/s41586-025-09761-x ---[CL] Real Deep Research for AI, Robotics and Beyond [UC San Diego & NVIDIA] https://arxiv.org/abs/2510.20809 ---[LG] Fluidity Index: Next-Generation Super-intelligence Benchmarks [QueueLab] https://arxiv.org/abs/2510.20636 ---[CL] Enterprise Deep Research: Steerable Multi-Agent Deep Research for Enterprise Analytics [Salesforce AI Research] https://arxiv.org/abs/2510.17797 ---[LG] The Free Transformer [FAIR at Meta] https://arxiv.org/abs/2510.17558

Ask this episode

“What did the council decide on item 42?” — answered with the speaker and the moment.

Transcript

Related