Patrick Cannon

I’m an independent AI researcher funded by BlueDot Impact. My current work addresses robust oversight for language-model reasoning. I study when learned verifiers become unreliable under inference-time search, how high-confidence false approvals occur, and when systems should defer rather than trust a verifier’s judgement.

I have a PhD in statistics from the University of Bristol, where I worked on particle MCMC for population genetics with Christophe Andrieu and Mark Beaumont.

At Improbable, I developed methods for calibrating large multi-agent simulations against real-world data. I later co-founded a computer-vision startup building large-scale 3D representations with neural radiance fields and Gaussian splats, before joining Amazon AGI to work on multimodal foundation models for speech and audio.