Crafting Effective Prompts
Prompts written for generative-AI tools are essentially computer code written in natural, everyday language. Natural words function like computer code. The more carefully you craft the prompt, the better the response (the “output”). As with so much in life: practice, practice, practice.
Start simply and iterate. See how changing words and adding more direction alter the results.
Specificity, simplicity and conciseness often generate better results. Usually, the more descriptive, specific, detailed and contextualized, the better. That said, you should also avoid unnecessary information and lack of precision. Don’t tell the tool what not to do; tell it affirmatively what to do. Aim for prompt language that is specific, simple and concise.
Remember that the prompt is in essence “computer code written in natural, everyday language.”
If it’s a large task you want it to perform, break it into sub-tasks. Also, for more complex tasks, after the stated task, try adding the prompt: “Let’s think step by step” to see how it affects the results.
To consider it formally, a prompt contains one or more of the following elements:
- Instruction – a specific task or instruction you want the tool to perform
- Context – additional information or context that can steer the model to better responses, e.g., telling the tool the output should be in the context of a university or design school or a particular type of course (you could even include the course description)
- Input Data – the task you want performed, e.g., to provide suggestions for a classroom activity or assignment
- Output Indicator – the type or format of the output, e.g., to write in the style of an 8th-grader or to simulate a university literature professor or undergraduate student in music composition
You do not need all the four elements for a prompt and the format depends on the task you want the tool to perform. Whether you are using prompts for generating text, images or code, the guidance is the same.
Some useful tips:
- Ask it to take on a role, e.g., tell it “You are a university professor teaching a first-year undergraduate course on Sustainable Systems Practices.”
- Give it context: you could add the course description for Sustainable Systems Practices and/or information about the students in the course, e.g., many students are English language learners, depending on the task you want performed.
- Be precise in what you’re looking for, e.g., “Your task: create two suggestions for an outdoor exercise with student groups that helps students make an emotional and intellectual connection to life forms in Union Square Park in New York City that are impacted by human intervention.”
See what ChatGPT did with this prompt.
Here are some additional resources:
Cynthia Alby: AI Prompts for Teaching
Ethan Mollick: Seven Approaches for Students with Prompts
Arturo Ozuna: General AI Prompting Guide for Educators
And if you really want to dive deep into prompt engineering, review this Prompt Engineering Guide from the organization Democratizing Artificial Intelligence Research, Education and Technologies (DAIR.AI)