Getting generative AI to provide references to its training data

The current generative AI systems such as ChatGPT are trained on existing texts, but they do not credit their sources. This project will develop a theoretical framework and novel methods for providing references to training data of large language models, which is a prerequisite to systems that are more transparent, trustworthy, and respectful of the rights of the original content creators. The grant will fund two Ph.D. students, one postdoc, and equipment.