Currently I am based in London, in the United Kingdom!

I'm in the MATS extension phase, having participated in the recent cohort in Berkeley California. My research there is focusing on understanding self-representation in LLMs, and whether self-representation might have important implications for safety. I'm currently mentored by Sid Black (UK AISI) and Oliver Sourbut (Future of Life Foundation).

I am also in the process of finishing up my master's thesis at the University of Cape Town, where I focused on trying to understanding how a neural network learns to solve a maze. We have a new paper up about that got into the NeurIPS Mechanistic Interpretability Workshop. See you in San Diego!

We also recently published our HumanAgencyBench paper, which was submitted to NeurIPS, accepted by the Area Chair and then cut due to space constraints (sad!).

I've also stepped back from being directly involved with the day to day operations of AI Safety South Africa, and am now in a more strategic role to help coordinate and support Leo Hyams. We are currently in the process of running the Cooperative AI Research Fellowship, which I think is a very exciting initative.

I also do weekly calls to help mentor people from the community and add value where I can. Sign up here if you're interested.