Unverified50% confidenceFactExact time
传统 LLM 将知识分布式编码在神经网络权重中（参数记忆），而 RAG 引入的外部文档存储属于可编辑的非参数记忆
1
Sources
50%
Confidence
Long-term
Relevance
7/6/2026
First Seen
Sources
从零构建RAG：用本地小模型打造领域专家系统
bilibili比特光锥_BightCone7/3/2026
Related Claims
UnverifiedRAG introduces external document storage as non-parametric memory, working in conjunction with parametric memory to address the knowledge cutoff date and hallucination problems of large language models.81% similarUnverifiedLLM网关提供请求频次控制、Token消耗统计、缓存命中、日志记录和成本预估等功能70% similarUnverifiedMem0 is positioned as a lightweight memory middleware whose core advantage lies in the memory management layer, automatically consolidating structured memories through an LLM-driven fact extraction and intelligent update mechanism.69% similarUnverifiedRAG通过引入外部知识库，将事实核查责任从模型参数记忆转移到可验证的文档来源69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/118900API
curl https://kongchang.com/api/v1/knowledge/claims/118900MCP
get_claim(id=118900)