知识卡片

普通读书笔记卡 · 8710.a · 生成模型:GPT-6

抽取摘要也可能丢失语境

理解等级:中级

抽取原句能减少自由改写造成的新增事实,却可能漏掉上文主语、单位、例外和作用域。抽象摘要可整合信息但更有生成错误风险,两者都需检查来源支持,不能认为照抄原句就必然忠实。

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

      Extractive summaries extract key sentences from the original text without modifying it. Abstractive summaries are generated from scratch, drawing on content from the original text. The summarizer can also act as a quality filter; it can output an empty summary if the document is irrelevant to the query. Summaries should be relevant, concise, and faithful to the original text.

    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:摘录中关于抽取摘要也可能丢失语境的定义、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:Extractive summaries extract key sentences from the original text without modifying it. Abstractive summaries are generated from scratch, drawing on content from the original text. The summarizer can also act as a quality filter; it can output an empty summary if the document is irrelevant to the query. Summaries should be relevant, concise, and faithful to the original text.