2 | Standard & Few Shot Prompts
2 | Standard & Few Shot Prompts
2024年6月19日创建
本文讨论了标准提示和少样本提示相关内容，介绍了标准提示输入大语言模型后会生成相应输出，还阐述了少样本提示的概念、示例及作用。关键要点包括：
1.
标准提示 ：向大语言模型（如ChatGPT）输入标准提示，模型会生成相应输出，例如输入“世界上最高的山是什么？”，输出为“珠穆朗玛峰”。
2.
少样本提示的概念 ：向模型展示少量（通常2到5个）示例，使其能快速适应先前遇到对象的新示例。
3.
少样本提示示例 ：如“彩虹的第一种颜色是什么？红色”“彩虹的第二种颜色是什么？黄色”“彩虹的第三种颜色是什么？” 以问答格式呈现。
4.
少样本提示的作用 ：将提示分解为少样本提示有助于构建和训练大语言模型，通过提供先前示例，更有可能获得想要的输出。
Let's take a quick look at standard or regular prompts. We'll go through how prompting a chatbot works and then quickly move onto few shot prompts.
Here's two regular prompt examples:
Regular Prompt
What's the highest mountain in the world?
Regular Prompt in Question and Answer format
Q: What's the highest mountain in the world?
A：
When you provide a regular prompt as an input to a large language model such as ChatGPT, it will generate a corresponding output or in prompt engineering terms, a completion. For example, this is illustrated below:
Q: What's the highest mountain in the world?
A: Mount Everest.
Now, there are also a few techniques you can employ to get better or more efficient outputs from a large language model such as Zero-shot, one-shot and few-shot prompting.
Few Shot Prompts
Few-shot prompting refers to a technique that involves presenting a small number of examples, typically ranging from two to five, to a model, enabling it to quickly adapt to new examples of previously encountered objects.