知识卡片
分步解题须比较收益与成本
分步生成可能为后续步骤提供中间结果,某些推理任务因此受益,但也会增加输出成本和引入错误传播。它不是对所有知识或分类任务都有效的通用开关;应以具体模型和任务评估,必要时用可验证工具处理算术。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 大语言模型入门
原书出处 · 1 处
- · 查看原书依据
Should we add the “think step-by-step” CoT instruction for every prompt, like a cheat code to a game? Sprague et al. evaluated CoT prompting over a wide variety of tasks and found that CoT primarily helps with tasks that need mathematical or logical reasoning. For tasks involving common-sense reasoning, they found that gains by CoT are limited. For knowledge-based tasks, CoT might even hurt.
参考来源
- 位置:《Designing Large Language Model Applications》第01章《大语言模型入门》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch01.html)
- 结论依据:摘录中关于分步解题须比较收益与成本的定义、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:Should we add the “think step-by-step” CoT instruction for every prompt, like a cheat code to a game? Sprague et al. evaluated CoT prompting over a wide variety of tasks and found that CoT primarily helps with tasks that need mathematical or logical reasoning. For tasks involving common-sense reasoning, they found that gains by CoT are limited. For knowledge-based tasks, CoT might even hurt.