[人人能懂] 揭秘顶尖模型的“偷懒”智慧

00:00:30 AI刷题也“偷懒”？聪明的懒办法，效率翻倍00:05:15 换个姿势“算命”，AI才能更聪明00:09:16 AI的省钱之道：从“专家会诊”到“超级记忆”00:13:08 组建梦之队：一个“鸟群智慧”给AI模型的启发00:16:55 AI界的“偏科生”：我们如何让它变成全能学霸？本期介绍的几篇论文：[LG] TreePO: Bridging the Gap of Policy Optimization and Efficacy and Inference Efficiency with Heuristic Tree-based Modeling  [ByteDance Seed]  https://arxiv.org/abs/2508.17445  ---[LG] Predicting the Order of Upcoming Tokens Improves Language Modeling  [MBZUAI]  https://arxiv.org/abs/2508.19228  ---[LG] UltraMemV2: Memory Networks Scaling to 120B Parameters with Superior Long-Context Learning  [ByteDance Seed]  https://arxiv.org/abs/2508.18756  ---[LG] PSO-Merging: Merging Models Based on Particle Swarm Optimization  [Chinese Academy of Science]  https://arxiv.org/abs/2508.19839  ---[CL] Demystifying Scientific Problem-Solving in LLMs by Probing Knowledge and Reasoning  [Yale University & Harvard University]  https://arxiv.org/abs/2508.19202  在小宇宙查看该单集文稿