About me

I'm interested in how neural networks represent and process information (mechanistic interpretability), and more broadly in questions about how to make AI systems trustworthy.

I'm currently on the Cognitive Oversight team at Anthropic. Before that, I was a founding research scientist at Goodfire, working on foundational questions about neural network representations.

I have always been drawn to using technical approaches to make sense of complex systems we don't fully understand. Before moving into AI safety, that meant within biomedicine and computational neuroscience. I'm originally from Australia, and moved to San Francisco at the end of 2022 to work on AI.