I am a third-year undergraduate student studying computer science and linguistics with a minor in korean studies at the University of Maryland.
I am fortunate to be advised by Professors Rachel Rudinger and Jordan Boyd-Graber.
Language models are increasingly capable, but our methods for measuring and building that capability lag behind. I work on evaluation and data pipelines for reliable NLP, namely:
1. Benchmark validity and the limits of what our evaluations actually measure
2. Human-AI collaboration in data annotation and curation
3. Evaluation for systems that reason, perceive, and act in the world
Nishant Balepur, Atrey Desai, Rachel Rudinger
While choices-only success is often deemed problematic, reasoning traces reveal that LLMs use less problematic strategies like inferring missing questions, challenging claims that partial-input success is always a flaw. Consequently, reasoning traces could help separate problematic data from less problematic reasoning.
Nishant Balepur, Bhavya Rajasekaran, Jane Oh, Michael Xie, Atrey Desai, Jordan Boyd-Graber
Atrey Desai, Sathvik Nair