3 papers
cs.CL2026
Prompt-Induced Score Variance in Zero-Shot Binary Vision-Language Safety Classification
Charles Weng, Dingwen Li, Alexander Martin
Safety pipelines often block content by thresholding a model's first-token unsafe score. We show that a single prompt's score is not a reliable probability and compare two repairs:…
cs.CV2025
HLTCOE Evaluation Team at TREC 2025: VQA Track
Dengjia Zhang, Charles Weng, Katherine Guerrerio +5
The HLTCOE Evaluation team participated in TREC VQA's Answer Generation (AG) task, for which we developed a listwise learning framework that aims to improve semantic precision and…
math.LO2025
More nonamalgamable forcing extensions
Miha E. Habič, Charles Weng, Cathy Zhang
We extend the results of arXiv:1808.01509 on nonamalgamable forcing extensions to families of posets with wide projections. We also use a different coding method to obtain nonamalg…