Research Interests
I am interested in interpretability research, specifically research targeting features of propositions (like attitudes held towards them, or reason-relations between them). My recent work along this line asks how LLMs internally represent truth values, and how those representations interact with in-context information during reasoning. I am also interested in developing new interpretability tools, with a focus on contrastive methods. In my research, I want to take seriously the questions that ask whether language models have, or can have, strong capabilities like understanding, reasoning, intentionality, etc. I look for precise operational definitions for such capabilities by drawing on philosophical accounts, and aim to test for them through mechanistic interpretability.