Agent memory is a critical part of agent architecture, and Stanford CS329Z goe… · @Michaelzsguo · TwKit

@Michaelzsguo

Agent memory is a critical part of agent architecture, and Stanford CS329Z goes deep into it. I learned a lot and can't wait to apply what I've learned to improve our own agent. So I turned Lecture 7, Agent Memory Architectures, into a 9-minute explainer. The central idea: An agent only sees what's in its context window, like a small pool of light in a dark room. Everything else is in the dark. Every memory system is essentially a way of deciding what to keep, where to store it, and when to bring it back into the light. What's inside: - 4 types of memory: Working, episodic, semantic, procedural - 4 design patterns: Voyager & skill files, MemGPT, Generative Agents, Mem0 - How memory is evaluated: LoCoMo → MemoryArena - What's still unsolved: What to forget, poisoned memories, over-personalization

曝光 176 · 评论 0 · 点赞 1 · 书签 2