[人人能懂] 乐高说明书、喜剧大赛与科研空间站

你有没有想过，AI的大脑里到底是什么样？今天我们就来一次深度探险，看看最新论文如何为我们绘制出AI的“乐高说明书”，又如何让它兼顾深思熟虑与脱口而出。我们还会把AI送上喜剧舞台，一探它那难以共情的奇特笑点，甚至把它放进一个虚拟“空间站”，看它能否成为真正的科学家。最后，我们会聊一个大趋势：AI正在悄悄地从遥远的云端，搬到你我的身边。

00:00:31 AI的“乐高”说明书

00:06:17 让AI既能深思熟虑，又能脱口而出

00:11:07 AI的笑点，为什么我们Get不到？

00:15:29 AI科学家，告别流水线

00:21:19 AI大变局：从云端到你身边

本期介绍的几篇论文：

[LG] Weight-sparse transformers have interpretable circuits  

[OpenAI]  

https://cdn.openai.com/pdf/41df8f28-d4ef-43e9-aed2-823f9393e470/circuit-sparsity-paper.pdf

---

[CL] TiDAR: Think in Diffusion, Talk in Autoregression  

[NVIDIA]  

https://arxiv.org/abs/2511.08923 

---

[CL] Assessing the Capabilities of LLMs in Humor: A Multi-dimensional Analysis of Oogiri Generation and Evaluation  

[Hitotsubashi University]  

https://arxiv.org/abs/2511.09133 

---

[LG] The Station: An Open-World Environment for AI-Driven Discovery  

[Dualverse AI]  

https://arxiv.org/abs/2511.06309 

---

[LG] Intelligence per Watt: Measuring Intelligence Efficiency of Local AI  

[Stanford University]  

https://arxiv.org/abs/2511.07885