Unverified50% confidenceTradeoffExact time
尽管Opus 4.7已有百万级上下文窗口，但上下文腐化（Context Rot）依然是巨大问题
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/18/2026
First Seen
Valid until: 10/16/2026
Sources
Related Claims
Unverified尽管上下文窗口很大，上下文腐化（Context Rot）依然是巨大问题，模型对上下文中不同位置信息的注意力分布不均匀（Lost in the Middle效应）74% similarUnverifiedContext Rot（上下文腐烂）指随着对话变长、输入 Token 增多，模型准确度会逐步下降，该现象在 Claude、OpenAI、Gemini 等主流模型上普遍存在62% similarUnverifiedRAG 系统中召回太多会浪费上下文窗口空间，召回太少则可能遗漏关键信息，检索精确度是关键挑战62% similarUnverifiedContext window limitations in LLMs cause them to struggle with schema information spanning hundreds of columns61% similarUnverified上下文腐烂（Context Rot）指随对话轮次增加，无关信息积累、注意力预算被低价值内容占满，导致推理质量持续下降61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/554704API
curl https://kongchang.com/api/v1/knowledge/claims/554704MCP
get_claim(id=554704)