Karen is interested in interpretability and faithfulness of LLMs. His research focuses on understanding and improving how these complex systems make decisions, aiming to make them more transparent and reliable.