Unverified50% confidenceFactExact time
Early SSMs performed limitedly on language tasks, mainly because their state update method was input-independent, making it difficult to selectively attend to information as flexibly as attention.
1
Sources
50%
Confidence
Long-term
Relevance
10/28/2026
First Seen
Sources
Related Entities
Related Claims
Verified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68% similarUnverified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65% similarUnverifiedCLAUDE.md is a soft constraint that influences large language model behavior through context injection, and the model may forget or ignore its instructions during long conversations due to context window limitations or attention drift.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1109750API
curl https://kongchang.com/api/v1/knowledge/claims/1109750MCP
get_claim(id=1109750)