Useful Memories Become Faulty When Continuously Updated by LLMs — Zhang et al.
直接挑战 LLM memory consolidation 的常见设计,给 Hermes 记忆系统提供了反证。
The site's interface is available in English, but most posts and notes are still in Chinese — translation is a work in progress.
Back
Original links I would recommend, each with a one-line reason.
直接挑战 LLM memory consolidation 的常见设计,给 Hermes 记忆系统提供了反证。
单用户 agent 场景下,搜索系统可以先做词法召回,再把理解交给 LLM。
让 prompt caching 从成本优化变成 agent prompt 组织约束,直接影响长期 loop 成本。
No curated resources match this filter.