主播
节目简介
来源:小宇宙
你是否也曾好奇,AI离拥有真正的“人性”还有多远?本期节目,我们将用几篇最新的论文,带你进行一次脑力激荡。我们会看到,如果用《帝国时代》里的绵羊也能搭建出一个AI,我们对它的“智能”判断是否会改变。接着,我们会探讨AI如何像我们一样学会“举一反三”,以及如何通过一套神奇的“外骨骼”,让机器人拥有人类的“手感”。最后,我们还会揭晓一种专为AI设计的“防作弊”考题,看看它到底是“原理型学霸”还是“题库型学霸”。准备好,让我们一起出发,探索AI认知的边界!
00:00:00 AI有灵魂吗?先问问“帝国时代”里的羊
00:06:03 AI的学习捷径,为什么“举一反三”比“死记硬背”更高效?
00:11:31 如何用“现在”的智慧,教会“过去”的自己
00:16:56 机器人摸着石头过河,靠的是什么?
00:21:50 你的模型,是真的懂了,还是在背题库?
本期介绍的几篇论文:
[CL] If LLMs Have Human-Like Attributes, Then So Does Age of Empires II
[Microsoft & The University of York]
https://arxiv.org/abs/2605.31514
---
[LG] Learn from your own latents and not from tokens: A sample-complexity theory
[EPFL & University of Cambridge & Johns Hopkins University]
https://arxiv.org/abs/2605.27734
---
[LG] Pretraining Recurrent Networks without Recurrence
[MIT]
https://arxiv.org/abs/2606.06479
---
[RO] Universal Manipulation Exoskeleton: Learning Compliant Whole-body Policies with Real-time Torque Feedback
[Ant Group]
https://arxiv.org/abs/2606.14218
---
[LG] A Held-Out Transition-Pair Falsifier for Long-Horizon Non-Abelian State Tracking
[Attractor Dynamics]
https://arxiv.org/abs/2606.07254
00:00:00 AI有灵魂吗?先问问“帝国时代”里的羊
00:06:03 AI的学习捷径,为什么“举一反三”比“死记硬背”更高效?
00:11:31 如何用“现在”的智慧,教会“过去”的自己
00:16:56 机器人摸着石头过河,靠的是什么?
00:21:50 你的模型,是真的懂了,还是在背题库?
本期介绍的几篇论文:
[CL] If LLMs Have Human-Like Attributes, Then So Does Age of Empires II
[Microsoft & The University of York]
https://arxiv.org/abs/2605.31514
---
[LG] Learn from your own latents and not from tokens: A sample-complexity theory
[EPFL & University of Cambridge & Johns Hopkins University]
https://arxiv.org/abs/2605.27734
---
[LG] Pretraining Recurrent Networks without Recurrence
[MIT]
https://arxiv.org/abs/2606.06479
---
[RO] Universal Manipulation Exoskeleton: Learning Compliant Whole-body Policies with Real-time Torque Feedback
[Ant Group]
https://arxiv.org/abs/2606.14218
---
[LG] A Held-Out Transition-Pair Falsifier for Long-Horizon Non-Abelian State Tracking
[Attractor Dynamics]
https://arxiv.org/abs/2606.07254