The KI-Chatbot ChatAI

Even if it is not advisable to copy the results and outputs of ChatGPT without checking them (see below under Implications), the chatbot can generally be used to generate ideas as part of the planning and design of courses and to save time when formulating texts or plans. For example, the following scenarios are possible:

• Creating seminar drafts: Based on a topic, create a draft for seminar and session content, and based on this, have announcement texts drafted for the course catalogue.
• Support for teaching planning: Design structures for semesters, seminars/lectures by integrating learning objectives and required teaching methods into the prompt. The chatbot can help to build the curriculum in a logical and coherent way.
• Creation of teaching materials: Prepare teaching materials such as PowerPoint presentations and handouts, e.g. enter topics and keywords and let ChatGPT generate suitable content.
• Customise seminar content: Recompile existing materials in different formats, enrich content or adapt tasks to different language levels.
• Develop exam questions: Develop tests, quizzes and open-ended questions, vary existing questions to minimise cheating attempts.
• Develop interactive learning activities: Design interactive learning activities such as role plays, debates and simulations using ChatGPT that encourage critical thinking and practical application of concepts.
• Support online discussion forums: ChatGPT can generate discussion topics and provide moderation input to ensure online forum discussions remain productive and support learning objectives.
• Provide feedback on student work: Provide initial feedback on student essays and reports, for example, to check grammar, style and coherence of content.
• Automate administrative tasks: Simplify administrative tasks such as creating student lists, scheduling classes or communicating with students by creating templates to free up time for teaching.

To integrate working with the chatbot directly into the classroom and encourage conscious use of ChatGPT, you could have students try out prompts and critique the results. Use ChatGPT to create questions based on your seminar/lecture slides for self-study by students. Work with students to establish rules for the use of AI in your seminar, e.g. where and under what conditions AI can and cannot be used.

Basically, the quality of the output depends on the accuracy of the prompt. Here are some tips for effective prompting:

• Start the same prompt in a new chat if the results of the first one were not satisfactory.
• Make your prompts as precise as possible.
• Give the chatbot contextual information and a role ("You are ...").
• Ask follow-up questions, e.g. "What else do you need to improve your answer?
• Structure the prompt, break tasks down and sequence them (build on each other).
• Encourage the chatbot to take extra care "because the result is important".

Save prompts that have worked well or that you want to use several times (slightly modified) so that you can copy them if necessary.
Use prompt catalogues that have already been set up by different institutions.

You should be aware of the following implications every time you use the chatbot, even though ChatAI now allows privacy compliant use:

• Copyrights, ancillary copyrights, moral rights are still not fully clarified - a good rule of thumb is to ask yourself: would I put what I want to put on ChatGPT on the internet with a clear conscience? It is therefore essential that you do not post any personal data or other sensitive data (research data, internal processes).
• ChatGPT is a language model, not a knowledge model - even if it looks like the AI thinks and knows. ChatGPT has been trained to complete input in a way that produces natural-sounding dialogues. For questions or tasks that the programme cannot actually handle, the chatbot 'hallucinates' terms until a coherent sounding text is produced.
• The training data used makes it possible to reproduce biases, such as using predominantly American data. ChatGPT's training data is not publicly available.

The output of the AI should always be checked - so you should consider whether it is still efficient to use, or whether checking it would take as much time and effort as working on a topic yourself.