Unverified50% confidenceFactExact time
In long conversations, earlier content may be truncated, sliding-window deleted, or summarized and compressed, causing key instructions or corrections to be lost.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在多轮对话场景中，随着历史消息累积，系统指令相对业务内容的比重下降，格式遵从度随之降低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71% similarUnverifiedExtractive summarization sacrifices some linguistic fluency and information compression rate in exchange for the verifiability that every sentence can be traced back to the source text.71% similarUnverifiedReal chat logs are filled with typos, memes, sentence breaks, and multi-party cross-references, and are bidirectionally mixed; training directly without cleaning will cause the model to learn chaotic patterns.70% similarVerifiedResearch shows that large language models exhibit a significant "Lost in the Middle" phenomenon when processing long texts, retaining information from the beginning and end of the input text well, but showing a marked decline in attention to information in the middle.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1037202API
curl https://kongchang.com/api/v1/knowledge/claims/1037202MCP
get_claim(id=1037202)