Unverified50% confidenceOpinionExact time
The causes of the lost-in-the-middle phenomenon are usually attributed to the decay effect of positional encoding, the fact that beginning and end content in training data is more often associated with answers, and the dilution of the attention softmax over long sequences.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
Verified研究表明当上下文过长时，模型对中间部分信息的注意力会显著下降，即"Lost in the Middle"现象76% similarUnverifiedContext confusion produces narrow misalignment, where misalignment is concentrated in specific domains semantically close to the training data, rather than a broad degradation of the model's overall behavior.76% similarUnverified模型在大海捞针测试中对中间位置信息的检索能力明显弱于首尾位置，这被称为注意力沉降现象75% similarUnverifiedAn "lost in the middle" effect exists in ultra-long contexts, meaning the model has stronger memory for content at the beginning and end of a sequence while attention to middle passages decreases; this phenomenon was named by Stanford researchers.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1029998API
curl https://kongchang.com/api/v1/knowledge/claims/1029998MCP
get_claim(id=1029998)