“Investigation of the Controllability of Large Language Models in a Dialogue System Context”

Preparatory Project (FNK)

Aims and Goals: The project focuses on improving the reliability and naturalness of interactions with dialogue systems implemented via large language models (LLMs). Key challenges include the controllability of LLMs, as seen in systems like ChatGPT. While such models generate grammatically correct outputs, the content can often be factually inaccurate due to training and optimisation methods that prioritise fluency over factual correctness.

Funding Period: 10/2024 – 09/2025

Contact: Prof. Stefan Ultes and Nicolas Wagner