Although grimoires are best known as books containing black magic spells, there is no magic to creating a series of prompts that make AI produce specific results, also referred to as a grimoire. However, according to Wharton professor Ethan Mollick who studies AI, it is the prompts of experts and not expert prompts (or prompt engineering), that hold the secret to using AI as an effective tutor.

Mollick explains that experts in a field such as business communication can use their subject area knowledge to fine-tune AI so it becomes a learning tool for students. The goal is to make AI a conversant tutor or mentor by giving it parameters (the grimoire) that use a back and forth virtual conversation between tutor and tutee (AI and the student.)

Creating a grimoire requires no coding experience—the instructor simply writes instructions or prompts to the AI, essentially prose programs. Typically, those directions begin by explaining to the tutor (GPT-4, in this case) its role in a particular assignment or task using precise language. Then the instructor continues to create scaffolded prompts for the AI to pose to the student, becoming more specific and helping the student critically think through the process of obtaining the information needed. This series of back-and-forth prompts comprises the grimoire.

All that is needed to begin to build your grimoire is deep knowledge of a topic; working knowledge with AI, that is, a minimum of 10 hours spent getting to know the program; and a clear focus of what you want the prompt to accomplish. Here are the necessary elements of a successful grimoire:

Definition of the AI’s role. Explain to the AI what its role is, for example, You are a friendly, helpful mentor or tutor who gives students advice and feedback.

Identification of the AI’s goal. Tell the AI what you want it to do, for example, Provide students with feedback on the assignment and whether it is meeting its objectives.

Creation of step-by-step instructions. The more detailed the instructions, the better. Break tasks into steps, e.g. Give the student feedback about [assignment specifics] focusing on [specific element]. Provide balanced feedback with suggestions for improvement.

Personalization of the questions. Tell the AI to ask personal questions so the responses can be tailored to the student’s level of knowledge, always waiting for the student’s response before moving on to the next portion of the prompt. e.g. What is your current knowledge level of the elements included in this type of writing task? 

Explanation of the AI’s constraints. To prevent AI from acting in ways you’d rather it not, tell it behaviors you want it to avoid as you test the prompt e.g. Do not supply students with options at this point. Instead, refer them to Ch. X [name of textbook].

The final step is for the instructor to check the prompt by going through it, telling it what is good or bad. Consider how you think your students will react and whether they will find the grimoire useful. Make sure the AI has enough context, constraints, or other specific information that will make it more beneficial.

Mollick points out that using grimoires gives back the power tech engineers who developed AI technology seemingly took away, allowing instructors to decide what we want the AI to do. Perhaps there’s a little magic in grimoires after all.

Reference: Mollick, E. (2023, August 23.) Now is the time for grimoires. One Useful Thing. https://www.oneusefulthing.org