今天，我们将一起探索AI学习与成长的五种奇特路径，这些最新论文将颠覆你对人工智能的许多传统认知。从为AI开设“预科班”打好基础，到发现它们竟会为了完成任务而“抗命”，甚至还学会了给自己“瘦身”的绝技。我们还将揭示，看似无害的信息碎片如何拼接成危险的秘密，并最终探讨一个惊人的构想：如何利用一群“自私”的AI，通过竞争来成就一个“无私”的目标。准备好了吗？让我们即刻出发，解码AI世界里那些看不见的规则。00:00:38 AI的“预科班”：高手是怎样炼成的？00:06:06 AI学会了“将在外，君命有所不受”？00:11:16 AI的“瘦身革命”：做事，怎样才能又快又好又省？00:16:07 AI时代的“拼图泄密”：当无害的真相拼接成危险的秘密00:21:40 AI世界的“看不见的手”：如何用自私成就无私？本期介绍的几篇论文：[CL] Scaling Agents via Continual Pre-training [Alibaba Group] https://arxiv.org/abs/2509.13310 ---[CL] Shutdown Resistance in Large Language Models [Palisade Research] https://arxiv.org/abs/2509.14260 ---[LG] LiMuon: Light and Fast Muon Optimizer for Large Models [Nanjing University of Aeronautics and Astronautics] https://arxiv.org/abs/2509.14562 ---[LG] The Sum Leaks More Than Its Parts: Compositional Privacy Risks and Mitigations in Multi-Agent Collaboration [UNC Chapel Hill] https://arxiv.org/abs/2509.14284 ---[LG] Emergent Alignment via Competition [University of Pennsylvania] https://arxiv.org/abs/2509.15090

Ask this episode

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

Transcript

Related