EPISODE · Oct 27, 2025 · 26 MIN
[人人能懂] 从算法自发现、流动性指数到可控AI
from AI可可AI生活
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 在小宇宙查看该单集文稿
Embed this episode
Ready to play
[人人能懂] 从算法自发现、流动性指数到可控AI
No transcript for this episode yet
Similar Episodes
No similar episodes found.