James Dixon
2nd year DPhil Student, Machine Learning Research Group, University of Oxford
About
I am an Oxford DPhil student in the Engineering Science department. My interests include AI safety and mechanistic interpretability. Supervised by Prof Stephen Roberts.
Running Commentary
- — New preprint: Steering by Influence: Curvature Aware Data Weighting for Activation Steering
We investigate whether model curvature information can be used to better weight more thematically relevant data for model steering.
- — Helping Organise a Neurips '26 Workshop
Foundations of LLM Post-Training in Changing Environments