Josh Yu
Computer Science & Psychology @ Yale University
Hey, I'm Huace but I usually go by Josh. I'm a first-year at Yale interested in AI safety and alignment. Right now I'm especially drawn to mechanistic interpretability and other ways of figuring out how models work from the inside.
Right now I'm researching paltering in AI models with Prof. Beth Anne Helgason at Yale SOM, exploring multi-paper reasoning in LLMs with Prof. Arman Cohan, and building Yale Meals as a developer at the Yale Computer Society.
Outside of that, most of my free time goes to music, movies, and tennis (I'm getting better at it).
Some of My Work
Detecting Paltering in AI Models
2026–presentResearch at Yale SOM on detecting when a model says only true things but still leads the user to believe something false.
Yale Meals
2026–presentOpen-source app for Yale dining menus and meal statistics, built with the Yale Computer Society and used by a large share of campus.
Deepfake Audio Detection
Summer 2026Pruning and retraining a transformer-based deepfake audio detection model for production, using whitening, early-exit tests and AUC probing.