Unverified50% confidenceFactExact time

RAG系统存在迷失在中间（Lost in the Middle）问题，即上下文较长时模型往往忽略中间部分信息而偏向开头和结尾

1

Sources

50%

Confidence

Long-term

Relevance

8/6/2026

First Seen

Sources

Alfa项目解析：用共振机制对抗AI幻觉的新思路

Related Claims

Verified模型对上下文窗口首尾内容的记忆显著优于中间部分，这一现象被称为「Lost in the Middle」问题78% similar VerifiedLLM存在'Lost in the Middle'现象，即模型对上下文开头和结尾的信息关注度高，而对中间部分的信息容易忽略77% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/695812

API

curl https://kongchang.com/api/v1/knowledge/claims/695812

MCP

get_claim(id=695812)