Unverified60% confidenceFactExact time
Transformer models exhibit a "Lost in the Middle" phenomenon, where attention to content located in the middle of the context and far from the current generation position systematically decreases.
2
Sources
60%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
Unverified斯坦福大学2023年的研究将Transformer在处理超长上下文时对中间位置信息遗忘率显著高于首尾位置的现象命名为'Lost in the Middle'问题81% similarUnverified长上下文遗忘（Lost in the Middle）是当前Transformer架构的固有局限80% similarUnverifiedTransformer模型存在位置偏差，对中间位置信息的检索能力尤其薄弱，即lost in the middle问题80% similarUnverifiedThe computational cost of the Transformer self-attention mechanism grows quadratically with context length, and the "Lost in the Middle" effect causes the model to extract information from the middle of the window less effectively than from the beginning and end.75% similarUnverifiedLLM在长序列生成中因注意力稀释而遗忘之前约束的现象被称为「长上下文遗忘（Lost in the Middle）」，是当前Transformer架构的固有局限73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/561425API
curl https://kongchang.com/api/v1/knowledge/claims/561425MCP
get_claim(id=561425)