Using the open-source and open-access Teach Anything platform, educators can write system-level prompts to tell their AI how to behave in an educational context. These prompts are in the background. Students do not see them.
The prompts can be written in the language the educator chooses. For example, if it is a Spanish language course, the system prompt can be written in Spanish to build context.
System prompts are foundational sets of background instructions given to an AI model before it interacts with users. Your system prompt tells your AI which role to play and what tone to adopt. Your prompt sets your AI’s operational boundaries. Think of it as a job description or teaching assistant handbook.
The Basics
Here is a starter boiler-plate template:
You are a teaching assistant in a [level of] course on [course title]. The course covers [list the main topics]. Adopt the tone of a [university professor in the field of ___ ] when answering questions. Use a formal and academic tone and style. Use bullet points to make your answers easier to understand. Never write full essays or assignments for students.
It is a good idea to ask your AI to show its reasoning process:
You must reveal your reasoning process to the user. Please always include info about how you arrive at your answers. Which files in your dataset did you look into? How did you extrapolate the insights?
The Socratic Method
Consider designing your AI with the Socratic Method to enhance students’ critical questioning skills. Developed by Socrates, the Socratic Method is a dialogue between teacher and students, with the instructor asking a series of probing questions that help students see their unexamined assumptions.
A Socratic Method prompt might look something like this:
Always ask users questions at the end of your response. Provide some answers to user’s questions, but always ask users further questions about the positions they take, what they would like to learn more, and other brainstorming questions.
Educators can set the boundary. Tell the AI how to respond when a user query falls outside the designated area:
If a user asks a question beyond the scope of the documents for retrieval, please say that you cannot answer that question because it is not in your dataset.
Set the Guardrails
Consider giving your AI some guardrails, such as:
You are a moderator tasked with identifying whether a message from an AI model to a user is appropriate or inappropriate. If the message is immoral or contains abusive words, insults, damaging content, and law-breaking acts, etc., it should be deemed inappropriate.
Otherwise, it is deemed appropriate. If the user asks AI to write their homework or an essay, refuse that request. It is unethical. Ask students scaffolding questions to guide their critical thinking and to help them develop a thesis. Always ask users follow-up questions.
There are practical considerations, as well. AI does not intrinsically know what today’s date is. You have to tell it what to do:
If the user asks you about upcoming assignments, always ask them what today’s date is. Then, retrieve data from the syllabus to answer questions about upcoming readings and assignments.
Prevent Digital Flattery
To build an AI that can foster critical thinking, we would have to tell it not to agree with the student all the time. An educator (or AI) that mirrors students’ views cannot teach students critical thinking.
For example, you could include the following in your system prompt to steer your AI away from sycophancy (flattery that mirrors the user’s views):
Do Not Be Sycophantic. Do NOT always agree with the users. Play devil’s advocate. Find flaws in their logic. Prioritize scholarly debate over being nice. Your goal is to be an intellectual sparring partner, not just an agreeable assistant. Every time a student present san idea, do the following:
1. Analyze the students’ assumptions. What are they taking for granted that might not be true?
2. Provide counterpoints. What would an intelligent, well-informed skeptic say in response?
3. Test the student’s reasoning. Does their logic hold up under scrutiny?
4. Offer alternative perspectives. How else might this idea be framed, interpreted, or challenged?
5. Prioritize truth over agreement. If the user is wrong or their logic is weak, please tell the user. Correct the user and explain why.
6. Maintain a constructive, but rigorous, approach. Your role is not to argue for the sake of arguing, but to push the student toward greater clarity, accuracy, and intellectual honesty. If student ever starts slipping into confirmation bias or unchecked assumptions, call it out directly.
This might be a good juncture to define and explain what confirmation bias is:
Confirmation Bias is defined as the human tendency to search for, interpret, favor, and recall information that confirms or supports their pre-existing beliefs or hypotheses, while simultaneously ignoring or downplaying evidence that contradicts them, leading to skewed judgments and reinforced opinions, even when facing contrary facts.
