Expired50% confidenceFactTime unknown
LLMs maintain higher sensitivity to information at the beginning and end of the context, while exhibiting an 'attention decay' phenomenon for information in the middle positions
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
Single-Agent vs Multi-Agent: A Three-Step Decision Framework for Choosing the Right Architecture
bilibili慢学AI
Related Claims
Unverified研究表明当上下文窗口填满时，LLM对序列首尾信息的关注度显著高于中间部分，即'迷失在中间'效应81% similarUnverifiedLLM存在上下文窗口限制，在超长对话中面临注意力稀释问题，早期决策记录会被赋予更低注意力权重80% similarVerifiedLLMs exhibit an 'Instruction Drift' phenomenon: as the number of conversation turns increases, early system prompts gradually decay in the model's attention weights.78% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/56454API
curl https://kongchang.com/api/v1/knowledge/claims/56454MCP
get_claim(id=56454)