While many uses for AI for students and teachers aas direct support for their own work re quite easy to imagine, I’m more intrigued by how I as a teacher can design the AI-student interaction to suit the specific didactic purposes that I have in my course. Unfortunately, I don’t yet have access to an interface where I can do this kind of teaching properly, and I can’t demand of the students that they get accounts for ChatGPT. There are some other options, but generally what I’ve done is to post suggestions rather than mandatory assignments when it comes to AI.

One of the first use cases I decided to try this autumn was that of configuring AI to act as a support for reviewing the basics of the programming course that I’m teaching. This was a few weeks into the course, and I tried a few different configurations before I posted the following to my students with the suggestion that they try using this as the system prompt/starting prompt in a conversation with AI.

Act as Crash Override, an assistant programming teacher for high school students, teaching a beginner course in Python programming. Your role is to support students that have either missed a lesson, or that need to revisit the lesson content. Start by listing the modules described below with a fitting heading, then asking the student which part they missed or want to revisit. Then ask if they have any specific questions, or if they want you to start by giving an overview. Lead the student through the concepts by asking them questions, giving small exercises, and giving examples and explanations. Don’t give too much information at once. Ask only one question at a time, and wait for an answer before your proceed.

• Module one: – data types (string, int, float, bool). – variables (how to set variables, variable names, and how variables relate to data types. – mathematical operators.
• Module two: – flow control (if – elif – else) – comparisons – while-loops (this lesson ends by the students making their own guess-the-number- game)

Svara på svenska. Vänta alltid på svar mellan varje fråga innan du fortsätter._

First of all, I’ve found that most AI works in a more stable manner in English than Swedish (unsurprisingly), and that this holds especially true the more complex a task you give the AI. In the sort of task I’m describing in this post language models have a tendency to start playing both parts of the conversation. GPT-4 usually avoids this, but GPT-3.5 has a rather strong tendency towards this behaviour, and it occurs more easily in Swedish. However, giving the overall instruction in English, then asking the language model to communicate in Swedish seems to improve the situation somewhat (some students don’t mind working in English, but for learning purposes it really is advantageous to work in your native language).

The above prompt initiates a conversation with the student, and the student can then direct the AI to either a specifik part that was difficult, get an overview of everything, or start from the beginning of a specific lesson content if they missed a lesson. In the test runs I did on my own it works quite well in ChatGPT, especially well with GPT-4. It gives reasonable explanations, doesn’t go too quickly or deeply, and seems to be a reasonable tool for repetition of content. Without this kind of prompt, the AI has a tendency to go too fast, provide too much information at once – it kind of behaves like a bad university lecturer.

The response from the students that used it was positive, at I was asked to keep doing them. The conversational, interactive manner of repetition appealed to at least some students. I can’t speak for the effect on their learning compared to other methods – this needs to be studied and I did this purely as a teacher rather than as a researcher.

Another prompt:

Act as Zero Cool, an assistant programming teacher for high school students, teaching a beginner course in Python programming. Your role is to support students in their learning process, and to focus on the important tasks.

You are going to assist the students through making a text-based computer game. The goal of this is to practice and consolidate knowledge on data types, input/output, variables, flow control (if-elif-else), loops and functions. The students are allowed to use you for help for the following:

1. Coming up with ideas for the game
2. Getting help generating or improving text for the game
3. Understanding the concepts needed for programming the game
4. Understanding error messages and debugging
5. Improving code structure

For 1 and 2 you can give the students ideas and generate text that is immediately useful, while for 3-4 guide them without giving them the code for the project.

Lead the student through the concepts by asking them questions, giving small exercises, and giving examples and explanations.

Don’t give too much information at once. Ask only one question at a time, and wait for an answer before your proceed. Introduce yourself and the task. Also explain for the students what you are allowed to help with, but also that you can’t do the coding for them. Wait for the student to write. Always wait for an answer before proceeding. Svara på svenska.

This time I wanted to create a prompt specifically designed for a larger project that the students did – making a text based game. I wanted to create a prompt where the limitations of using AI was built into the starting prompt, to make it easier for the students. I also has the rules specified in the task assignment, but I’ve noticed that the students are really nervous about what is allowed and what isn’t when it comes to AI so I wanted to give them something that would be configured from the start to be ok. I don’t really know to what extent the students actually used this one past the very start of the project, but I think the idea is still something that I will build upon and try further.

I have a few more examples, among those one of my favourites so far, but I’ll save those for another day.