本期我们将拆解5篇最新论文，带你见证AI如何打破常识：它能在零数据的“小黑屋”里顿悟万物规律，也能在认知空间里把复杂决策化为优雅的“两点一线”。你会看到聪明的机器人如何靠看懂“事物关系”实现“看一眼就会”，而大模型狂揽长文本的“免费午餐”幻觉也终于被现实戳破。更不可思议的是，机器竟学会了量化“品味”，开始自主打捞真正有趣的真理宝藏。00:00:33 把AI关进小黑屋，它竟自己顿悟了世界的底层逻辑？00:05:19 别再盲目死磕，AI学会把复杂难题“两点一线”，给了普通人什么启示？00:09:30 看一眼就会的真本事，从“死记硬背”到“举一反三”的底层逻辑00:14:54 别被大模型的长篇大论骗了，AI的“免费午餐”为何走到尽头？00:20:36 当人工智能有了“品味”，如何在无限的信息中寻找真正的宝藏？本期介绍的几篇论文：[AI] Self-Play Pretraining with Zero Data[Tel Aviv University & Stanford University]https://arxiv.org/abs/2609.30063---[RO] Representation World Model: Learning States, Transition and Executable Plans in Representation[Tsinghua University]https://arxiv.org/abs/2609.29171---[RO] RAPID: Robot Agentic Programming from Demonstrations[MIT & University of Pennsylvania & National University of Singapore]https://arxiv.org/abs/2609.30249---[CL] No More Free Lunch: Corpus Task Complexity Matters as Corpora Grow[UC Berkeley & CMU]https://arxiv.org/abs/2609.2924---[LG] Learning to Discover Interesting Mathematics[Meta]https://arxiv.org/abs/2609.28603在小宇宙查看该单集文稿

Ask this episode

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

Transcript

Related