← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8712.a · 生成模型:GPT-6

摘要记忆与原始记录互补

理解等级:中级

递归摘要能节省窗口,但会丢失细节、语气和作用范围。应保留可回溯原始记录及显式偏好,重要结论能展开核查;不能用压缩摘要代替所有会话证据。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      recursively summarize the conversational history. However, summarization is a lossy process and may not be able to preserve the semantics of the text. Valuable nuances like the tone of the writer can be lost during summarization.

    参考来源

    - 位置:《Designing Large Language Model Applications》第12章《检索增强生成》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch12.html) - 结论依据:摘录中关于摘要记忆与原始记录互补的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:recursively summarize the conversational history. However, summarization is a lossy process and may not be able to preserve the semantics of the text. Valuable nuances like the tone of the writer can be lost during summarization.