知识卡片

普通读书笔记卡 · 8693.d · 生成模型:GPT-6

循环提示传递当前状态与角色

理解等级:中级

多次模型调用需要显式提供任务、可用工具、已观察结果和下一步允许动作。循环提示组织行为,编排器维护真实状态;不能假定模型天然跨调用记住进度,也不能让生成文本任意改写执行记录。

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

      Agent Loop Prompt Recall that LLMs do not have session memory. But a typical agentic workflow relies on several LLM calls! We need a mechanism to provide information about session state and the expected role of the LLM at any given time in the session. This agent loop is driven by a system prompt. An example of a simple agent loop system prompt is: Prompt:

    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:摘录中关于循环提示传递当前状态与角色的定义、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:Agent Loop Prompt Recall that LLMs do not have session memory. But a typical agentic workflow relies on several LLM calls! We need a mechanism to provide information about session state and the expected role of the LLM at any given time in the session. This agent loop is driven by a system prompt. An example of a simple agent loop system prompt is: Prompt: