Hi, I'm a research scientist on the alignment team at the UK's AI Security Institute (AISI). I mainly focus on how complexity theory can be used in AI alignment, which typical involves work on AI safety via debate.
I recieved my PhD from Leiden University for my contributions to quantum computing, in topics from quantum complexity theory to quantum machine learning and circuit cutting, where I was advised by Vedran Dunjko. My personal favourite from this line of research is a proof that if classical computers can compute all functions that quantum computers can, then the polynomial hierarchy collapses to it's second level (with some light additional assumptions).
Alongside my PhD I completed a few projects in AI safety: One using information theory to analyse how models may encode information and how interpretable these encodings would be with Jan Kirchner and I worked with Beth Barnes on the first round of what became METR Evals, assessing the risk posed by modern LLMs as part of MATS 2.0.
Before my PhD I was aligned as an undergraduate in Maths+Physics at Warwick University, where I completed my Master dissertation in fluid dynamics under James Sprittles . Through this connection, I can trace my mathematical genealogy back to Poisson, Cauchy, Lagrange, Laplace and Euler. I have an Erdős number of 4 through a collaboration with Scott Aaronson and a Bacon number of Infinity that I don't get from a collaboration with Jack Nicholson. I am open to any collaboration that would lower either of these numbers.
Contact
Relevant contact information can be acquired via email.