Unverified50% confidenceFactExact time
即便是支持128K上下文的模型，在文档长度超过32K时，中段信息的检索准确率往往出现明显下滑
1
Sources
50%
Confidence
Long-term
Relevance
7/17/2026
First Seen
Sources
Related Claims
UnverifiedIn document chunking strategies, chunks that are too long reduce retrieval precision, while chunks that are too short result in loss of context.74% 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.71% similarUnverifiedEven large models with a 128K token context window experience attention dilution when processing large amounts of history.69% similarUnverifiedEven large language models that support context lengths of 128K or longer will experience attention dilution when all capability instructions are injected at once, causing a decline in execution quality for specific tasks.69% similarUnverifiedThe capacity of human working memory is approximately 7±2 chunks of information (Miller's Law), and when system complexity exceeds this threshold, the comprehensibility of plain-text code drops sharply.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/545095API
curl https://kongchang.com/api/v1/knowledge/claims/545095MCP
get_claim(id=545095)