I am a PhD candidate at the University of Michigan School of Information, advised by Eric Gilbert and Ceren Budak.
We introduce an evaluation framework that tests whether LLMs learn fundamental human values or merely surface-level preferences. Across 9 models, the average Deep Value Generalization Rate is just 0.30, meaning all models generalize deep values less than chance.
We introduce Plurals, a system and Python library for pluralistic AI deliberation. Three randomized experiments show simulated focus groups produced output resonant with relevant audiences (chosen over zero-shot generation in 75% of trials). Python Library | Docs
We conducted an experiment (800+ participants, 40+ countries) using a dynamic "many-worlds" design. We find that high AI exposure increased collective diversity but not individual creativity. AI made ideas different, not better.
I create generative art using programming, randomness, and geometry. My work explores the intersection of mathematical patterns and aesthetic expression.