My research objective is to build machines that plan and learn with language. People use language to plan in their everyday lives: imagine constructing a piece of furniture with guidance from the instruction manual, cooking with somebody else in the kitchen, or correcting a child's mistakes building an erector set. My goal is to understand how people efficiently learn from language instructions, as well as represent their actions and intentions through language, and build machines that work in similar ways.
My current work has focused on building and analyzing benchmarks for decision-making tasks that require planning and learning from language (AI Agents that Matter, CORE-Bench), as well as designing AI systems that leverage language understanding to learn action abstractions and better plan (Ada, Goal Inference). For my senior thesis, I am studying what mechanisms people use to predict the goals of others around them by running controlled human experiments and building computational models for how they might make predictions.