Hi, I’m Mark

I studied math and computer science at The University of Chicago. My academic research has been largely in LLM interpretability., in particular I’m interested in a comprehensive approach: we should study LLMs at both the mechanistic and behavioral level, using our findings in a complementary way to make models more interpretable, useful, and human-aligned. I currently work with Victor Veitch and Ari Holtzman at UChicago, and Micah Goldblum at Columbia. In the past I’ve worked in astrophysics, NLP, AI for science, and quantitative social science. I’ve loved computers for a long time, getting my start building a mail server in my basement in elementary school. I spend a lot of my free time in other parts of the stack—collecting analog electronics and building custom PCBs and devices. Is everything computer? Beyond computers I love modernist architecture from the 20th century, riding and repairing my bike (I am NOT a bike guy though—lycra is humiliating and should be permanently banned), and organizing with my community. I’m very into public policy, social science, and the humanities (wait, did I just say I’m interested in everything?). I was a national circuit public forum debater in high school, have a Twitter account where I post a little too much, and am always getting into a slightly embarrassing philosophical kick (currently Heidegger, formerly Foucault). I guess I should say I’m into movies, but any real film heads please do not take this so seriously.