Researchers propose a novel internal hidden state framework for detecting hallucinations in large language models, achieving improvements in precision-recall AUC. They also introduce a cross-model detection framework where one model observes another model's internal representations to detect hallucinations.