[人人能懂] AI的“小灶、泥人与神级助理”

00:00:29 解锁AI推理能力：少即是多的秘密00:05:17 AI巨兽的“成长配方”：如何让小个子智慧指导大块头？00:10:03 AI作画的“临门一脚”：如何让快马跑得又快又好？00:14:05 AI 创造 3D 模型：从“盖房子”到“捏泥人”00:07:44 AI界的“小灶”，如何喂出绝顶高手？本期介绍的五篇论文：[LG] Part I: Tricks or Traps? A Deep Dive into RL for LLM Reasoning  [Alibaba Group]  https://arxiv.org/abs/2508.08221 ---[LG] μ-Parametrization for Mixture of Experts  [University of Warsaw]  https://arxiv.org/abs/2508.09752 ---[LG] Noise Hypernetworks: Amortizing Test-Time Compute in Diffusion Models  [Technical University of Munich & Inceptive]  https://arxiv.org/abs/2508.09968 ---[CV] VertexRegen: Mesh Generation with Continuous Level of Detail  [Meta Reality Labs Research & UC San Diego]  https://arxiv.org/abs/2508.09062 ---[LG] Beyond Scaling Law: A Data-Efficient Distillation Framework for Reasoning  [Zhongxing Telecom Equipment (ZTE)]  https://arxiv.org/abs/2508.09883 ---[CL] Speed Always Wins: A Survey on Efficient Architectures for Large Language Models  [Unknown / Survey paper]  https://arxiv.org/abs/2508.09834 在小宇宙查看该单集文稿