Unverified85% confidenceOpinionTime unknown
RAG（检索增强生成）的召回质量直接决定Agent的决策质量，检索失败意味着后续所有推理都建立在错误基础上
1
Sources
85%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedContext contamination caused by RAG produces 'retrieval-driven hallucinations,' and such errors typically do not trigger exceptions or error codes and can only be detected through manual semantic verification.72% similarUnverifiedThe answer quality of RAG depends largely on the retrieval step rather than on the model's capabilities themselves.71% similarUnverifiedRAG (Retrieval-Augmented Generation), fact-checking chains, and confidence scoring are multi-layered defense mechanisms against LLM hallucinations.70% similarVerifiedRAG has drawbacks including unstable retrieval precision, semantic matching deviations, and consumption of large amounts of the context window.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/20179API
curl https://kongchang.com/api/v1/knowledge/claims/20179MCP
get_claim(id=20179)