Project C07 investigates how large language models (LLMs) can be trained, adapted and used within a retrieval-augmented generation (RAG) setting to co-construct explanations together with users instead of just generating them. The focus is on AI systems recognizing in dialogue what has already been understood, where uncertainties still exist, and how explanations can be built up step by step.

To this end, computer scientists are developing new methods that enable AI systems to capture the course of a conversation and respond to it in a targeted manner. On this basis, relevant specialist knowledge and important contextual information are dynamically selected and incorporated into the explanation. This ensures that the language model always takes the current context into account in its explanations.

Research areas: Computer science

Project duration: from 01/2026