AI的效率革命：当“更巧”比“更大”更重要

00:01:45 AI 炼丹术：发现模型训练的“万能公式”？ 00:07:01 AI军备竞赛，弹药快打光了怎么办？ 00:12:16 让 AI 学会“反思”，会发生什么？ 00:16:48 AI“高考”也偏科？换个考法，学霸变学渣 00:22:08 AI的“外挂”：当你的短板，成了我的跳板 00:27:18 让AI“看人下菜碟”：下一代人工智能的效率革命 本期介绍的六篇AI前沿论文：[LG] Scaling Collapse Reveals Universal Dynamics in Compute-Optimally Trained Neural Networks  [Google DeepMind & New York University]  https://arxiv.org/abs/2507.02119 ---[LG] Fast and Simplex: 2-Simplicial Attention in Triton  [Meta & University of Texas at Austin]  https://arxiv.org/abs/2507.02754 ---[LG] Energy-Based Transformers are Scalable Learners and Thinkers  [Amazon & UVA]  https://arxiv.org/abs/2507.02092 ---[CL] Answer Matching Outperforms Multiple Choice for Language Model Evaluation  [Max Planck Institute for Intelligent Systems & Tübingen AI Center]  https://arxiv.org/abs/2507.02856 ---[RO] MultiGen: Using Multimodal Generation in Simulation to Learn Multimodal Policies in Real  [UC Berkeley]  https://arxiv.org/abs/2507.02864 ---[LG] Reasoning on a Budget: A Survey of Adaptive and Controllable Test-Time Compute in LLMs  https://arxiv.org/abs/2507.02076 在小宇宙查看该单集文稿