Unverified50% confidenceFactExact time
Context pollution refers to a large language model's context window being occupied by irrelevant system prompts, unused tool descriptions, or redundant conversation history, which compresses the token space available for reasoning and leads to degraded instruction-following quality, confused tool selection, or unstable outputs.
1
Sources
50%
Confidence
Long-term
Relevance
9/22/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedContext Degradation refers to the phenomenon in which large language models gradually lose the ability to retrieve and utilize early information as input content accumulates.74% similarVerified大语言模型在长时间对话中会因上下文窗口限制而丢失早期信息73% similarUnverifiedCLAUDE.md is a soft constraint that influences large language model behavior through context injection, and the model may forget or ignore its instructions during long conversations due to context window limitations or attention drift.73% similarUnverifiedLarge language models exhibit an 'instruction drift' phenomenon, where the model's probability of following early instructions gradually decreases as conversation turns increase and the context window is filled with large amounts of code.70% similarUnverified大型语言模型在无约束空间中会倾向于创造性发散，而非工程所需的收敛执行70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/941049API
curl https://kongchang.com/api/v1/knowledge/claims/941049MCP
get_claim(id=941049)