Unverified50% confidenceOpinionExact time
RAG检索依赖语义向量相似度匹配，无法检索只存在于PR评论或工程师记忆里的隐性决策知识
1
Sources
50%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
LangChain开源OpenWiki：自动为代码库生成并维护AI可读Wiki
bilibiliGoldenSpiderAI7/3/2026
Related Claims
UnverifiedRAG-based memory retrieval has an inherent limitation: retrieval granularity is constrained by text chunking strategy, and similarity matching cannot perfectly capture complex causal reasoning chains and temporal dependencies73% similarUnverifiedRAG系统中检索结果通常按相关性分数排序后顺序拼接，最相关的chunk不一定排在最显眼位置70% similarUnverifiedRAG（检索增强生成）、事实核查链和置信度评分是对抗LLM幻觉的多层防御机制70% similarUnverifiedRAG系统的检索模块主要依赖语义相关度而非信源可信度筛选文档，导致语义相关的假新闻可能比语义稍弱的权威报道获得更高检索优先级68% similarVerifiedRAG技术通过将问题转为向量检索最相近知识片段并拼入提示词，可大幅减少大模型幻觉68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/163559API
curl https://kongchang.com/api/v1/knowledge/claims/163559MCP
get_claim(id=163559)