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
VerifiedLLMs exhibit an 'Instruction Drift' phenomenon: as the number of conversation turns increases, early system prompts gradually decay in the model's attention weights.80% similarUnverifiedLLMs maintain higher sensitivity to information at the beginning and end of the context, while exhibiting an 'attention decay' phenomenon for information in the middle positions80% similarUnverifiedOverly long system instructions in LLMs lead to 'attention decay' on critical rules due to non-uniform attention distribution across the context window.78% similarUnverifiedProofreading generated content in a new session can avoid the attention decay problem caused by an excessively long context window.77% similarUnverifiedAs the number of conversation turns increases, the weight of early key constraint information in the context window is diluted (attention drift), causing the quality of AI output to gradually decline.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)