Unverified50% confidenceFactExact time
In RAG systems, long documents are split into chunks that are independently encoded and stored in a vector database; chunks separated from their context may lose critical information (e.g., pronouns may fail to accurately capture semantics).
1
Sources
50%
Confidence
Long-term
Relevance
10/5/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDocuments retrieved in a RAG system are concatenated into the context window, and the model often cannot distinguish trusted system instructions from untrusted external data.78% similarUnverifiedThe information input boundary of RAG is locked in at the retrieval stage; if key information is missed during the retrieval step, no matter how powerful the subsequent generation stage is, it cannot compensate.75% similarVerifiedRAG (Retrieval-Augmented Generation) retrieves relevant document chunks from a vector database and injects them into the prompt before calling the LLM; when the retrieved documents are mismatched or contain incorrect information, 'retrieval-driven hallucinations' can occur.74% similarUnverifiedRAG将文档切成固定大小chunk，跨文档、跨时间的逻辑关联在切片时被破坏，向量相似度检索无法重建这些隐性联系74% similarUnverifiedSplitting documents into fixed-length segments in RAG leads to low retrieval accuracy; documents should instead be chunked by semantic boundaries (paragraphs, sections, logical units) to ensure each chunk is a complete semantic unit.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/975246API
curl https://kongchang.com/api/v1/knowledge/claims/975246MCP
get_claim(id=975246)