Unverified50% confidenceFactExact time
LLM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
1
Sources
50%
Confidence
Long-term
Relevance
7/17/2026
First Seen
Sources
Related Claims
VerifiedLLM存在'Instruction Drift'现象：随着对话轮次增加，早期系统提示在模型注意力权重中逐渐衰减80% similarUnverifiedLLM对上下文开头和结尾的信息保持较高敏感度，而对中间位置的信息存在'注意力衰减'现象80% similarUnverifiedOverly long system instructions in LLMs lead to 'attention decay' on critical rules due to non-uniform attention distribution across the context window.78% similarUnverified在新会话中对生成内容进行审校可避免上下文窗口过长导致的注意力衰减问题77% similarUnverified随着对话轮次增加，上下文窗口中早期关键约束信息权重被稀释（注意力漂移），导致AI输出质量逐渐下降77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/544766API
curl https://kongchang.com/api/v1/knowledge/claims/544766MCP
get_claim(id=544766)