This repository aims to provide a practical and intuitive way to use prompts based on knowledge and requirements. The scope of this guideline is general, but it primarily uses the development process as an example. Additionally, the approach is demonstrated through examples involving GitHub Copilot.
This quadrant approach is straightforward. First, it's necessary to understand that most problems you encounter when using LLM chats involve two controlled factors:
- Knowledge about the topic: How much you know about the topic addressed in the question.
- Requirements for the question's objective: Does your question aim to solve or respond to something specific, and do you have all the necessary requirements for an efficient response?
These two factors can be visualized in a quadrant chart, creating four main zones. Each zone corresponds to different approaches in prompt engineering.
quadrantChart
title Quadrant for Prompting Approach 🤖
x-axis Low knowledge --> High knowledge
y-axis Unclear Requirements --> Clear Requirements
quadrant-1 Zero/One shot
quadrant-2 Full Prompt
quadrant-3 Multi Prompt
quadrant-4 Reverse Prompt/CoT
The details of the quadrant and the reasons behind each approach can be explored further in the following section.
-
Base knowledge for prompt engineering patterns
This section covers foundational patterns in prompt engineering, explaining how each pattern works and its benefits. -
Quadrant for prompt engineering
This section introduces a quadrant framework to help select the appropriate prompting approach based on knowledge level and requirement clarity. -
Examples and uses
This section provides practical examples illustrating how to identify the appropriate quadrant and prompting approach for different scenarios.
Important
It's important to better understand prompt engineering patterns, making the first section the recommended starting point