Research on how large language models represent beliefs, goals and uncertainty, combining interpretability, behavioral and agent-based evaluation, probabilistic and cognitive modeling to extract, evaluate and manipulate internal representations with high confidence.