I’m interested in how neural networks represent and process information (mechanistic interpretability).

Currently, I’m a founding research scientist at Goodfire, working on foundational questions about neural network representations. I get a lot of joy from my work! It is intrinsically fascinating to me and I also believe it matters a lot. I believe interpretability is crucial for building AI systems we can trust and understand. I also believe that we can use interpretability to uncover new things about our world (e.g. novel scientific discoveries). Broadly, AI has overwhelming potential to shape an amazing future (if we get it right!).

Current Interests

1. Feature geometry & manifolds.
2. Possible connections between adversarial examples and superposition.
3. Barriers to scaling interpretability techniques.

Background

I’m originally from Australia, moving to San Francisco at the end of 2022 for a startup I cofounded. I am a med school dropout. Prior to moving to the US, my research was broadly using technical skills to gain insight into biomedicine (e.g. computational neuroscience, computer vision for radiology).

Reach Out!

I love meeting people who share my curiosity about AI, interpretability, or just interesting ideas in general.

Email: liv[at]livgorton[dot]com

Twitter: DMs open!