Unverified50% confidenceFactExact time
上下文窗口大小不等于有效上下文长度，超出有效注意力范围会导致遗忘漂移现象
1
Sources
50%
Confidence
Long-term
Relevance
7/10/2026
First Seen
Sources
GLM-5.2发布：开源编程能力首次逼近闭源旗舰模型
bilibili五里墩茶社6/22/2026
Related Claims
Unverified更长的上下文窗口并未消除注意力稀释效应，反而带来更高的单次调用成本79% similarUnverifiedProofreading generated content in a new session can avoid the attention decay problem caused by an excessively long context window.76% similarUnverifiedEven for large models that support 128K tokens, attention decay is a practical concern in actual use—the model's attention to information in the middle of the window is significantly lower than at the beginning and end.75% similarUnverifiedWhen prompts are too long, large models experience an 'attention explosion' phenomenon, where excessively long inputs cause effective information to be lost.75% similarUnverified当前大模型并没有真正的持久化记忆机制，随着对话轮次增加，早期需求描述在注意力机制中的权重会逐渐衰减75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/438700API
curl https://kongchang.com/api/v1/knowledge/claims/438700MCP
get_claim(id=438700)