Advait Deshmukh

Hi! I’m Advait, a Computer Science Ph.D. student at the University of Colorado Boulder, advised by Dr. Maria Antoniak. I work on language technology and its responsible use. I’m interested in the inductive biases these systems rely on, and — just as much — in the assumptions people make about what they’re capable of, as individual users and as a society. Understanding both sides is how we locate the gaps between them, and closing those gaps is what leads to better outcomes: for the science, and for the people it touches. So far, I’ve studied the limitations language models inherit from their training distributions (All Entities are Not Created Equal), and how people actually engage with them for story generation, using naturally occurring conversations from WildChat (The Garden of Forking Prompts).

Within these broader themes, I keep returning to narrative and discourse. Partly that’s personal — I love stories and novels, and I think stories are an essential part of human experience; any technology that sets out to “help” people will do it better for having learned from that. Partly it’s that discourse is genuinely messy: concepts like coherence and diversity resist clean definition, and in an era of reinforcement learning with verifiable rewards, the irreducible subjectivity of narrative is exactly what makes it a compelling problem to work on.

I’m part of the CLS Lab at CU Boulder. I previously worked at — and continue to collaborate with — the BLAST Lab, led by Dr. Maria Pacheco.

I spent this past summer as an Applied Science Intern at Amazon in Bellevue, on the long-term planning and forecasting team within Supply Chain Optimization Technologies (SCOT), where I improved evaluation for their agentic systems. Previously, I worked for two years as a software engineer at Société Générale in Bangalore, where — among other things — I built a retrieval-augmented chat system that nudged me toward the NLP research I do today.

For fun, I like reading books, playing chess, and painting cat-sitting.