Unverified50% confidenceFactExact time
Vector space is essentially dimensionless in time; it can only assess semantic similarity between texts and cannot understand which memory is more recent or has become outdated.
1
Sources
50%
Confidence
Long-term
Relevance
9/3/2026
First Seen
Sources
Related Entities
Related Claims
Unverified大语言模型在生成查找目标文本时依赖记忆重建而非逐字符复制，在长上下文场景下会出现记忆漂移导致字符串替换静默失败，此现象被称为幻觉性精确71% similarUnverified更大的上下文窗口并不等于长期记忆，它只在单次推理中能处理更多内容，但不会在会话之间保留信息69% similarVerifiedSemantically similar texts are also close in distance within vector space, which is the mathematical foundation of vector retrieval.68% similarUnverifiedPlain text prompts are inherently ambiguous when describing precise spatial control, as language is discrete and abstract while image editing requires continuous and precise spatial control.67% similarUnverifiedText chunking granularity that is too large results in retrieved text containing large amounts of irrelevant information, while granularity that is too small may lose contextual semantics.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/851117API
curl https://kongchang.com/api/v1/knowledge/claims/851117MCP
get_claim(id=851117)