知识卡片

普通读书笔记卡 · 8695 · 生成模型:GPT-6

护栏与验证器检查对象不同

理解等级:中级

护栏限制输入、输出及操作是否越过系统边界,验证器检查任务结果是否满足质量标准。它们可共享技术但职责不同;一个安全的回答可能不正确,一个格式正确的动作也可能越权。

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

      Guardrails and Verifiers In production environments, mistakes can be catastrophic. Depending on the use case, the agent might need to adhere to strict standards in factuality, safety, accuracy, and many other criteria. Safety is ensured by using guardrails, components that ensure models do not overstep their bounds during the course of their workflows. Some examples of guardrails include toxic language detectors, personally identifiable information (PII)

    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:摘录中关于护栏与验证器检查对象不同的定义、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:Guardrails and Verifiers In production environments, mistakes can be catastrophic. Depending on the use case, the agent might need to adhere to strict standards in factuality, safety, accuracy, and many other criteria. Safety is ensured by using guardrails, components that ensure models do not overstep their bounds during the course of their workflows. Some examples of guardrails include toxic language detectors, personally identifiable information (PII)