Investigate the state of the art in digital watermarking in the symbolic domain. In particular, conduct literature research as well as practical experiments in order to find out whether or not LLM answers (eg by ChatGPT) can be manipulated invisibly (eg by changing fill words) so that such answers can later be recognized as originating from an LLM. Requires good knowledge in natural language processing, data science and signal processing. Open for single students and groups of two students.
