Unverified50% confidenceFactExact time
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
1
Sources
50%
Confidence
Long-term
Relevance
8/5/2026
First Seen
Sources
Related Claims
UnverifiedA 2023 Stanford University paper titled 'Lost in the Middle' found that models pay significantly less attention to information in the middle of the context window than at the beginning and end, exhibiting a U-shaped attention distribution.75% similarUnverifiedThe root cause of attention dilution lies in Softmax normalization: the more tokens in a sequence, the effective attention weight allocated to each position decreases approximately inversely proportional to the sequence length.75% similarUnverifiedCurrent frontier models face attention decay when handling long-range dependencies, with the ability to maintain consistency of early-established settings declining significantly as context length increases.74% similarVerified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74% similarUnverifiedHaving too many tokens in a prompt dilutes the attention mechanism's ability to focus, reducing the actual influence of each description.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/692299API
curl https://kongchang.com/api/v1/knowledge/claims/692299MCP
get_claim(id=692299)