As artificial intelligence (AI) continues to gain popularity, it is essential for members of our campus community to develop skills in prompt literacy. To effectively utilize AI models, you should understand the capabilities and limitations of these models, and be able to write specific and detailed prompts. By building and refining these skills, you can effectively integrate AI into your work and education while mitigating many problems and frustrations that may arise.
Prompting for Education Purposes
In addition to the resources below, more in-depth training is available in the Generative AI Prompt Literacy Canvas course. The prompt literacy information below is also available as a one-page poster, provided by U-M Flint.
- Understand the Capabilities and Limitations of AI
AI models, such as GPT-3.5 and 4, are powerful, but they have limitations. At this time, no GenAI technology actually ‘understands’ the thing it’s writing about in the way a human does. Natural language processing chatbots do not know what a cup of coffee is, they just use probability to select the correct words to write extensively about it. Additionally, they can't provide opinions or make value judgments, and they can't access real-time or personal data unless it's included in the prompt. - Learn About the Domain You’ll be Prompting
As you write your prompts you’ll need an understanding of the topic or domain you're pointing the AI toward. This will help you with many of the following recommendations, but also account for things like hallucinations (outputs that sound plausible, but are incorrect or unrelated) and misinformation. You’ll also be able to give the AI more context and precision in your query. Finally, the more you know about the domain you’re writing for the more likely you’ll know if the output is accurate and of high quality. - Be Specific and Detailed
By default, AI will give broad and general responses. The more specific and detailed your prompt is, the more useful the AI's response is likely to be. If you're asking the AI to generate text, provide as much context as possible. If you're asking a question, make sure it's clear and unambiguous. Be sure to indicate:- The desired goal of your content.
- The desired language and tone that is appropriate for your audience.
- The specific format you want for the output.
- The focus of your theme for the content, e.g. nutrition, history, analytics, engineering.
- Minimize Logic ‘Work’
Going hand-in-hand with the ‘be specific’ advice above, you want to minimize the number of assumptions and logical reasoning the GenAI will need to make on its own to create your content. The tool will have a limited amount of processing resources available. A more detailed prompt that makes each of the AI tasks more bite sized will yield better performance and higher quality output. - Be Creative with your Prompt
Prompting a GenAI requires a collaborative approach to problem-solving with that AI. To achieve your best results, you should think creatively and use a variety of conversational approaches when developing your prompt. The output you have formulated in your mind likely won’t match what the AI algorithm will produce. Sometimes you’ll need to think/write like an expert negotiator, debater, or sales person to get the output you have in mind. - Use Iterative Refinement
It is unlikely that you will get a “perfect” response on your first try. Don't be afraid to refine your prompt and try again. You can also use a technique called "iterative refinement," where you provide feedback on the AI's responses and ask it to improve them. Many prompt sessions evolve into lengthy dialogues. - Tailoring to your Audience
As you evaluate and refine your prompt, make sure the output is appropriate for your audience and adjust as needed. Not only do you need to make sure the content is accurate and makes sense for the topic, but also appropriate for the reader's knowledge level. Don’t be afraid to tell your GenAI who the target audience is, specifically. - Consider Ethical Implications
Be mindful of the ethical implications of using AI in education. This includes issues like privacy, fairness, and transparency. Always make sure stakeholders understand when and how AI is being used. - Stay Up-to-Date
AI is a rapidly evolving field. Stay up-to-date with the latest research and developments to understand how to best use AI in your work.
Best Practices
Below you will find a recommended list of best practices that will accelerate your learning and understanding of prompt writing for improved output when working with generative artificial intelligence. This is not an exhaustive list.
Adjustments
- Tone
- Length
- Temperature
- Voice (e.g. enthusiastic, empathetic, optimistic, authoritative...)
- Avoid confusing words that can have multiple meanings, if possible.
- Request alternative or varied output if what’s produced does not meet expectations.
Be Specific
- Instruct GenAI to ask follow-up questions to improve the output or clarify confusion. It's okay to ask the AI if it understands what you're asking.
- When developing for the web/Canvas, instruct it to structure the results in a table/markdown/HTML.
- Provide more details if the output is ambiguous.
- Give examples of what you're looking for.
- To reduce the likelihood of hallucinations (outputs that sound plausible, but are incorrect or unrelated), tell it that it's an expert in the topic you're prompting. For example, "You are an expert in men's fashion design." Other expert role examples include:
- Copywriter
- Public Speaker
- Marketing Strategist
- UX Designer
- SEO Expert
- Programming Tutor
- Web Development Expert
Setting Boundaries
- Point out any mistakes it makes.
- Tell it what you do and don't like for output.
- Instruct the GenAI to stay on topic.
- Repeat important instructions as necessary.
- Describe what it should and should not write about.
Prompting Tasks
- Use sequential prompting: Break up your prompts by informing it that you'll provide additional input later.
- Have GenAI inform you when it's ready for that input.
- Stop the output and rephrase your query if the output is not what you expect. A common example is "Make it shorter and less formal."
- Request justification as needed. Ask "Why did you write that?"
Prompt Examples
Reference the examples developed by Flint ODE that produced desired output for additional guidance in your work with GenAI. Refer to the U-M GenAI Prompt Library for a comprehensive resource designed to support the diverse needs of the University of Michigan community.