Unverified80% confidenceFactTime unknown
当LLM上下文窗口中的指令过多时，模型对每条指令的关注度会下降，导致遵循率降低
1
Sources
80%
Confidence
Long-term
Relevance
6/2/2026
First Seen
Sources
Related Entities
Related Claims
Unverified随着指令数量增加，LLM的遵循质量会整体下降，开始无差别忽略所有指令而非仅忽略靠后的指令84% similarUnverified研究表明LLM在处理过长、过复杂的指令集时整体遵循率会系统性下降，即指令遗忘现象80% similarUnverifiedOverly long system instructions in LLMs lead to 'attention decay' on critical rules due to non-uniform attention distribution across the context window.76% similarUnverifiedCurrent model instruction-following failures commonly manifest as ignoring system prompt constraints, forgetting early instructions in long conversations, randomly violating formatted output requirements, and deviating from predefined processes in multi-step tasks.76% similarUnverified指令衰减（Instruction Decay）是指当上下文窗口逐渐被填满时，模型对早期系统指令或用户要求的遵循程度明显下滑73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/37623API
curl https://kongchang.com/api/v1/knowledge/claims/37623MCP
get_claim(id=37623)