Hi, I’m Tom! I am a second year PhD candidate in machine learning (ML) supervised by Prof. Philip Torr (University of Oxford) and Dr. Tim G.J. Rudner (NYU). My general research interests concern AI safety. In particular, how can we elicit, produce and give guarantees for the safe and reliable deployment of ML models, specifically LLMs and variants such as reasoning models.

I have several publications at top-tier ML conference including ICML, NeurIPS and ICLR. These include papers on: the semantic calibration of LLMs; steering LLMs adaptively at inference time; biases in VLMs; and giving formal statements on universal in-context approximation of fully recurrent models including SSMs.

In the past, I used to a be a little better at mathematics. You can find some related material and work in the resources section.

Feel free to reach out to chat or discuss anything ML or mathematics related!