你有没有想过，能写诗作画的AI，为什么有时却像个固执的孩子？本期我们要聊的几篇最新论文，就试图教会AI一些我们习以为常、但它却难以理解的人类智慧。我们将一起看看，如何治好AI的“路痴”症，让它拥有空间感；如何让它从被动看图，变身主动破案的“侦探”；甚至，如何通过巧妙的“换个姿势”，让它终于听懂“不要”，并随心所欲地调整观察事物的“粒度”。00:00:33 人工智能的“路痴”难题00:05:24 AI侦探，如何给千米大桥做“体检”？00:09:59 从“你猜”到“你定”：AI图像分割的新玩法00:14:45 换个姿势，让AI听懂“不要”本期介绍的几篇论文：[CV] Scaling Spatial Intelligence with Multimodal Foundation Models [SenseTime Research] https://arxiv.org/abs/2511.13719 ---[CV] BridgeEQA: Virtual Embodied Agents for Real Bridge Inspections [University of Houston] https://arxiv.org/abs/2511.12676 ---[CV] UnSAMv2: Self-Supervised Learning Enables Segment Anything at Any Granularity [UC Berkeley] https://arxiv.org/abs/2511.13714 ---[CV] SpaceVLM: Sub-Space Modeling of Negation in Vision-Language Models [MIT] https://arxiv.org/abs/2511.12331

Ask this episode

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

Transcript

Related