ritwikg [at] umd.edu
Assistant Professor, University of Maryland
Technical Director for Autonomy, Defense Innovation Unit (DIU)
Wan-Cyuan Fan, Jiayun Luo, Declan Kutscher, Leonid Sigal, Ritwik Gupta
VLMs are selectively blind. They decide how much to look at an image based on question framing (open-ended vs Yes/No vs MCQ), even when the same visual reasoning is required. We analyze this phenomenon and mitigate it in this paper.
Milton Gomez, Marie McGraw, Saranya Ganesh S., Frederick Iat-Hin Tam, Ilia Azizi, Samuel Darmon, Monika Feldmann, Stella Bourdin, Louis Poulain-Auzéau, Suzana J. Camargo, Jonathan Lin, Dan Chavas, Chia-Ying Lee, Ritwik Gupta, Andrea Jenney, Tom Beucler
TCBench is a benchmark for evaluating global, short to medium-range (1-5 days) forecasts of tropical cyclone track and intensity. It builds on the IBTrACS observational dataset and includes state-of-the-art dynamical and neural weather models. Designed for accessibility, TCBench helps AI practitioners tackle domain-relevant TC challenges.
Ritwik Gupta, Andrew W. Reddie
U.S. AI security policy is increasingly shaped by an LLM Mirage, the belief that national security risks scale in proportion to the compute used to train frontier language models. That premise fails in two ways: it miscalibrates strategy because adversaries can obtain weaponizable capabilities with task-specific systems, and it destabilizes regulation because compute thresholds are easily renegotiated as domestic priorities shift.
