I’m grateful to be part of Open Philanthropy’s Century Fellowship, which supports people early in their careers to work on challenges the world may face this century that could have a lasting impact on the long-term future.

I’m interested in the concern that incautious deployment of powerful but poorly-understood deep learning models could result in global catastrophe in my lifetime. The default outcome of “successfully” scaling up existing ML training techniques may be systems that pursue proxy goals, disconnected from their intended purpose – going on to deceive humans and building influence until the situation is irrecoverable.

After dismissing this possibility as far-fetched for several years, I now think it is worth taking seriously. You may be interested to read more here or here, or feel free to contact me with questions.

Previously, I completed my DPhil at the University of Oxford, supervised by Tom Melham and Daniel Kroening, where my research focused on the testing and evaluation of state-of-the-art image classification models.

Before that, I taught for two years at a comprehensive secondary school as part of the Teach First Leadership Development Programme, and studied computer science at the University of Cambridge.