Abstract:

In this paper, we present the assistant feature of ChatGPT on the Q&A platform https://frag.jetzt. We show how AI assistants can be used to create abstracts, glossaries, Anki flashcards, mind maps, quizzes, and exam questions. These are based on course-specific materials such as uploaded scripts, book chapters, slides, images, data collections or transcripts. The paper focuses on answering questions "extractively" with the help of an AI assistant, i.e. the answers are generated from the provided content and cited with the source: "Retrieval-Augmented Generation". The aim of the RAG technique is to avoid hallucinations – misleading or incorrect information – that can occur with generative language models without specialist knowledge.

Another focus is the contextualization of AI assistants through "pre-prompting". This involves preparing the context prior to the interaction so that the bot provides personalized responses, e.g. in line with the learning content or desired skills. This enables tailored support, for example by suggesting specific learning materials based on the learner's skill profile.

As part of an evaluation study, AI assistants are available free of charge to teachers and students on frag.jetzt. The service is anonymous and does not require an OpenAI account. We encourage teachers and students to embark on an enlightened and safe journey into the world of AI-assisted education that prioritizes privacy and is free from misinformation.

Keywords:

Artificial Intelligence, ChatGPT, Educational Technology, Retrieval-Augmented Generation, Contextualization Techniques, Data Security, Hallucination Mitigation.