I’m a Ph.D. candidate at Stanford CS, advised by Chelsea Finn. My research is supported by OpenAI and KFAS.

My research focuses on operationalizing text as a substrate for learning. As tasks grow more complex, low-bandwidth scalar signals can’t keep up. These require learning from higher-bandwidth feedback that preserves the structure of what went wrong. I develop methods that enable models to extract massive amounts of information from direct experience through structured textual feedback such as natural-language corrections, pairwise comparisons with “why better” explanations, and reasoning traces.

Rather than treating text as throwaway scaffolding, I view it as a persistent store to optimize, where models accumulate experience at increasing levels of abstraction, similar to how humans write papers and books. This combines parametric models (for inductive biases and in-context understanding) with nonparametric text storage (for persistence and interpretability). Looking forward, I’m focused on scaling these methods to scientific discovery and other open-ended domains that require continual learning across long horizons.

Recent papers along these lines: