Unverified50% confidenceFactExact time
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
1
Sources
50%
Confidence
Long-term
Relevance
7/24/2026
First Seen
Sources
Related Claims
Unverified传统对话历史模式的局限性包括：无法原生表达工具调用的中间状态、缺乏时序元数据、不支持并发事件流、在上下文窗口有限时难以进行选择性压缩和摘要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74% similarUnverifiedIn long conversations, historical messages are repeatedly included in the context, causing the token consumption per interaction to grow continuously; the actual billing for the 10th round of dialogue includes the full text of the preceding 9 rounds.72% similarUnverifiedThe cascaded architecture has a flaw: each stage introduces latency, and the intermediate text conversion loses paralinguistic information such as tone, emotion, and speech rate.72% similarUnverifiedIn multi-turn agent conversations, token consumption is often 5–10 times that of ordinary conversations due to the need to carry tool call results, historical context, and system prompts.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/604384API
curl https://kongchang.com/api/v1/knowledge/claims/604384MCP
get_claim(id=604384)