[人人能懂] 从“野蛮生长”到“精准调校”：AI幕后的五位“工匠”

00:00:37 AI界的乐高革命：如何让你的模型“活”在当下00:05:41 AI的“手术刀”：我们如何精准“切除”它的坏心思？00:10:22 机器人学功夫：抄近道，还是练笨功？ 00:15:03 AI怎么当学徒：让机器学会看“领导脸色” 00:20:18 给AI动个“小手术”，治好它的“选择困难症”本期介绍的几篇论文：[LG] SequenceLayers: Sequence Processing and Streaming Neural Networks Made Easy  [Google DeepMind]  https://arxiv.org/abs/2507.23292  ---[LG] The Geometry of Harmfulness in LLMs through Subconcept Probing  [Algoverse AI Research]  https://arxiv.org/abs/2507.21141  ---[LG] Retrieve-Augmented Generation for Speeding up Diffusion Policy without Additional Training  [The University of Tokyo]  https://arxiv.org/abs/2507.21452  ---[LG] NPO: Learning Alignment and Meta-Alignment through Structured Human Feedback  [Microsoft & Amrita Vishwa Vidyapeetham]  https://arxiv.org/abs/2507.21131  ---[LG] TokenBlowUp: Resolving Representational Singularities in LLM Token Spaces via Monoidal Transformations  [University of Washington]  https://arxiv.org/abs/2507.19747   在小宇宙查看该单集文稿