Unverified50% confidenceFactExact time
When generating longer structured outputs, large language models experience type degradation due to the attention mechanism's decaying weight on early context—for example, serializing an entire array as a string or writing numbers as strings.
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
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78% similarUnverifiedLarge language models encode textual statistical patterns into parameters via backpropagation and do not store raw data in the traditional sense, but high-frequency text may produce memorization phenomena and be reproduced nearly verbatim.75% similarUnverifiedLarge language model text generation is an autoregressive serial process that produces only one token at a time, making it impossible to fully parallelize the output stage, with higher computational intensity than the input encoding stage.73% similarUnverifiedLarge language models form parametric memory by compressing and encoding training corpora into weight matrices, which is distributed rather than exact storage, and this is one of the reasons for hallucinations about brand information.73% similarUnverifiedThe output of large language models is fundamentally probabilistic text generation, and a single rephrasing of a prompt can change field names, nesting levels, or enumeration values in the output without triggering compile-time or static analysis warnings.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/831567API
curl https://kongchang.com/api/v1/knowledge/claims/831567MCP
get_claim(id=831567)