← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6760.a · 生成模型:GPT-6

护栏应区分内容类别与事实错误

理解等级:中级

歧视、隐私、越界话题和事实错误有不同判据,单一坏话标签难覆盖全部。可组合多标签分类、规则、检索核验等方法,并按场景设阈值;某类毒性检测通过不等于内容真实或动作已获授权。

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

      There is a virtually unlimited number of things you don’t want your chatbots to say. Most of these messages can be classified into one of two broad categories: toxic or erroneous. Here are some examples of some toxic messages your NLP filters will need to detect and deal with. You should be familiar with some of these aspects of toxicity from the toxic comments dataset

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第10章《真实世界中的大语言模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-10.html) - 结论依据:摘录中关于护栏应区分内容类别与事实错误的定义、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:There is a virtually unlimited number of things you don’t want your chatbots to say. Most of these messages can be classified into one of two broad categories: toxic or erroneous. Here are some examples of some toxic messages your NLP filters will need to detect and deal with. You should be familiar with some of these aspects of toxicity from the toxic comments dataset