← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8675.a · 生成模型:GPT-6

知识与自知需分别评估

理解等级:中级

模型是否能回答,与是否知道自己答不了,是两个不同问题。可用可回答和不可回答问题检验正确回答、恰当弃答及无依据断言;知识四象限只是分析框架,不能由模型口头自称知道就确定它有相应知识。

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

      Known knowns The LLM knows this knowledge/skill and is able to utilize it. Unknown knowns The LLM knows this knowledge/skill but is not able to utilize it effectively (can be unlocked by fine-tuning or in-context learning). Known unknowns The LLM knows that it does not know this knowledge. Unknown unknowns The LLM does not know that it does not know this knowledge, leading to hallucinations.

    参考来源

    - 位置:《Designing Large Language Model Applications》第08章《对齐训练与推理》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch08.html) - 结论依据:摘录中关于知识与自知需分别评估的定义、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:Known knowns The LLM knows this knowledge/skill and is able to utilize it. Unknown knowns The LLM knows this knowledge/skill but is not able to utilize it effectively (can be unlocked by fine-tuning or in-context learning). Known unknowns The LLM knows that it does not know this knowledge. Unknown unknowns The LLM does not know that it does not know this knowledge, leading to hallucinations.