[人人能懂] 安全、记忆与效率的新范式

00:00:27 AI的安全围栏，为什么总有漏洞？00:04:26 AI记性不好？我们该如何给它补补脑00:08:39 猫鼠游戏：我们如何给越来越聪明的AI装上“紧箍咒”？00:14:12 AI的“直觉”：它知道答案，只是还没说00:18:57 AI世界的“群策群力”：如何用三个臭皮匠，干翻一个诸葛亮本期介绍的五篇论文：[LG] On Surjectivity of Neural Networks: Can you elicit any behavior from your model?  [UC Berkeley]  https://arxiv.org/abs/2508.19445  ---[CL] Memory-R1: Enhancing Large Language Model Agents to Manage and Utilize Memories via Reinforcement Learning  [Ludwig Maximilian University of Munich & Technical University of Munich]  https://arxiv.org/abs/2508.19828  ---[LG] Reliable Weak-to-Strong Monitoring of LLM Agents  [Scale AI]  https://arxiv.org/abs/2508.19461  ---[CL] Diffusion Language Models Know the Answer Before Decoding  [The Hong Kong Polytechnic University & Dartmouth College & Max Planck Institute for Intelligent Systems]  https://arxiv.org/abs/2508.19982  ---[LG] Symphony: A Decentralized Multi-Agent Framework for Scalable Collective Intelligence  [Emory University & Columbia University]  https://arxiv.org/abs/2508.20019  在小宇宙查看该单集文稿