Unverified50% confidenceFactExact time
Locomo（Long-Context Memory）侧重评估跨会话的上下文连贯性与时序推理
1
Sources
50%
Confidence
Long-term
Relevance
9/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLoCoMo和LongMemEval等对话回忆基准测量的是对对话历史的问答能力，而非记忆是否改变了工具使用智能体的实际行为76% similarUnverified短期记忆一般保留最近5-10轮对话的原始内容，并在超出阈值时触发压缩总结71% similarUnverified生产级 agent 需要区分短期记忆（当前会话上下文）和长期记忆（跨会话的用户偏好、历史事实）70% similarUnverified长程记忆评测集（如 LongMemEval、MemGPT 基准）考察五个维度：单跳信息提取、跨会话推理、知识更新、时间推理、拒答判断68% similarVerified梳理思维导图的过程是一次主动回忆（Active Recall），效果优于被动翻看笔记66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/947379API
curl https://kongchang.com/api/v1/knowledge/claims/947379MCP
get_claim(id=947379)