About Me
This is my space for thinking through ideas at the edges of what I understand. This space is ever changing and I have found that it is better to be open about my beliefs and uncertainties.
My Work
My background is in NLP and ML. I studied how language models generalise across domains at the Univeristy of Cambridge(BA) and the University of Edinburgh (MSc), built evaluation frameworks for AI systems at LinkedIn, and trained in technical AI safety (mechinterp, RL, evals) at ARENA.
More recently, I did the ERA fellowship where I worked on two projects. Firstly, I worked on the governance mechanisms for Chain of Thought Monitorability and the potential for targeted transparency disclosures. Secondly, I worked on how models can have different linear representations for traits depending on the persona that they are role-playing as. I am now a technical governance research manager at ERA and a mentor for the MARS fellowship. Additionally, I have been doing some strategy consulting for the Centre for Long Term Resilience on their Loss of Control Observatory.
More generally, I try to look at the full landscape of AI safety work and the strategy for the space going forward. This includes thinking about the technical safety work and implications for governance but also the talent pipeline and the funding ecosystem.
What You'll Find Here
- •Essays — Writing on AI, technology, and related ideas
- •Atlas — My current takes on AI safety topics
- •Code & Projects — Technical experiments and tools