PhillipsLab — University of Utah

Kahlert School of Computing · Led by Prof. Jeff M. Phillips

Interpretable AI · Activation Steering · Dataset Curation

📊 Featured Datasets

Conciseness ↔ Verbosity Contrast — minimal pairs isolating conciseness vs verbosity for interpretability and activation steering.
Positivity ↔ Negativity Contrast — stance-style pairs controlling positive vs negative sentiment while keeping factual parity.