Now

I'm currently a MATS 9.0 fellow, based in sunny Berkeley! I'm working on the science of evaluations with Alex Turner. In particular, I want to figure out how we can make safety and scheming propensity evals still faithfully predict behaviors of interest as models become increasingly evaluation-aware. It would be a real shame if pre-deployment evals stopped working, and I want to prevent that.

I'm also co-organizing Proxima for winter 2026 AI safety fellows in the Berkeley community with my friend Parv, and the Idealists Unconference in the UK this April for the wonderful Idealists Collective.

I've also been thinking a lot lately about gradual disempowerment and civilizational resilience.

Current goals

• Finishing up Gray Swan research developing better web agent evaluation infrastructure
• Learning salsa! Reach out if you want to go to weekly Friday salsa classes in SF together