Verified65% confidenceFactExact time

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

3

Sources

65%

Confidence

Long-term

Relevance

7/11/2026

First Seen

Sources

GPT-5.6实测：近千行代码一次搞定，无需反复提示

twitterOpenAIDevs7/10/2026

Related Claims

UnverifiedRAG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78% similar Partially Verified学术界将大模型倾向于关注输入文本开头和结尾、忽略中间部分关键信息的现象称为'Lost in the Middle'问题76% similar Unverified大模型的记忆仅限于当前上下文窗口内的token序列，对话结束后所有中间状态消失75% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/488236

API

curl https://kongchang.com/api/v1/knowledge/claims/488236

MCP

get_claim(id=488236)