你有没有想过，我们该如何真正地“驾驭”AI？本期节目，我们将深入AI的“引擎室”，从五篇最新论文出发，探讨几个迷人的问题：我们应该怎样把开发AI应用从手工作坊升级为高效的“流水线”？AI能像我们一样拥有一个高效的“专家委员会”和灵活的记忆吗？抛开模仿，我们能否给AI“捏”出一个真实的性格？甚至，AI会不会是我们从未谋面的“认知表亲”？最后，我们又该如何用“廉价”的数据，教会机器人办成“昂贵”的事？00:00:35 你的AI应用，该升级“作坊”为“流水线”了00:07:42 AI进化启示录，从“大力出奇迹”到“聪明地长大”00:13:08 AI是我们的“远房表亲”吗？00:21:17 我们能给AI“捏”出一个人格吗？00:28:28 机器人教练，怎样用“廉价”的数据，办成“昂贵”的事？本期介绍的几篇论文：[AI] What makes prompts a graph: necessary and sufficient conditions for prompt graph engineering[Federal Institute of Goiás]https://arxiv.org/abs/2607.27578---[CL] Kimi K3: Open Frontier Intelligence[Kimi Team]https://arxiv.org/abs/2607.24653---[AI] Cognitive Convergence: Deep Similarities Between Large Language Models and Human Cognition[University of Michigan]https://arxiv.org/abs/2607.26179---[CL] From Representations to Behaviors: Exploring the Person-Situation-Behavior Triad in LLMs[Peking University & Beijing Institute of Technology]https://arxiv.org/abs/2607.26853---[RO] HiFi-UMI: Learning Deployable Manipulation Policies from High-Fidelity UMI Data Alone[Simple Al]https://arxiv.org/abs/2607.25895在小宇宙查看该单集文稿

Ask this episode

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

Transcript

Related