你有没有想过，聪明的AI也需要精打细算？本期节目，我们就来聊聊AI世界里的那些“增长智慧”：如何像果蝇大脑一样“聪明地偷懒”，又如何像请了私教一样精准地突破瓶颈。我们还会探讨，AI究竟应该把知识背下来还是学会查资料，以及机器人怎样才能在漫长任务中给自己“打气”加油。这些最新论文里的奇思妙想，不仅关乎技术，更藏着我们都能借鉴的策略。00:00:32 AI省钱的终极奥义，深度思考，一次缓存00:05:29 AI养成记，喂知识，还是给书单？00:12:23 如何让机器人学会“干大事”？给它一个好报酬，再加一个好心态00:18:31 你的大脑偷懒，可能比你想象的更聪明00:24:31 AI卡壳了怎么办？请个“私教”来帮忙本期介绍的几篇论文：[CL] Universal YOCO for Efficient Depth Scaling [Microsoft Research] https://arxiv.org/abs/2604.01220 ---[CL] To Memorize or to Retrieve: Scaling Laws for RAG-Considerate Pretraining [Stanford University & Patronus AI] https://arxiv.org/abs/2604.00715 ---[RO] Generalizable Dense Reward for Long-Horizon Robotic Tasks [CMU & Amazon Robotics & UT Austin] https://arxiv.org/abs/2604.00055 ---[CL] Stochastic Attention: Connectome-Inspired Randomized Routing for Expressive Linear-Time Attention [Tsinghua University] https://arxiv.org/abs/2604.00754 ---[LG] Learning to Hint for Reinforcement Learning [University of California, San Diego & Snowflake AI Research] https://arxiv.org/abs/2604.00698

Ask this episode

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

Transcript

Related