知识卡片

普通读书笔记卡 · 8696.b · 生成模型:GPT-6

蕴含检验只能验证来源支持

理解等级:高级

把源文作前提、摘要句作假设,可用NLI判断是否受支持。它检验的是源文忠实性,不自动保证源文真实;模型还可能受长文本、否定、数量及跨句推导限制,需要目标域校准。

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

      we do not have access to ground truth. But for summarization applications, we do have access to the ground truth. Therefore, we can verify factuality by taking each sentence in the summary and checking whether, given the source text, one can logically conclude the statement in the summary. This can be framed as a natural language inference (NLI) problem, which is a standard NLP task.

    参考来源

    - 位置:《Designing Large Language Model Applications》第10章《外部工具与智能体》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch10.html) - 结论依据:摘录中关于蕴含检验只能验证来源支持的定义、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:we do not have access to ground truth. But for summarization applications, we do have access to the ground truth. Therefore, we can verify factuality by taking each sentence in the summary and checking whether, given the source text, one can logically conclude the statement in the summary. This can be framed as a natural language inference (NLI) problem, which is a standard NLP task.