Hello,I am Linlin Wu, a first-year PhD student.
I study how the geometry of learned representations can make language models more interpretable and controllable.
Research
Representation geometry, large language models, and AI alignment.
I am interested in when the internal spaces of neural networks are curved, distorted, or otherwise resistant to simple linear descriptions—and what that means for reliably steering model behavior.
Selected preprint · 2026
Curveball Steering: The Right Direction To Steer Isn’t Always Linear
Shivam Raval*, Linlin Wu*, Hae Jin Song, Abir Harrasse, Jeff M. Phillips, Fazl Barez, Amirali Abdullah
* Equal contribution
Read on arXiv ↗Background
Mathematics has shaped how I think about structure and learning systems.
PhD in Computing
University of Utah · 2025—present
MS in Mathematics
Northeast Normal University · 2022—2025