If people did not sometimes do silly things, nothing intelligent would ever get done.
Ludwig Wittgenstein
Hello! I'm Ryan, an empirical safety researcher interested in alignment and mechanistic interpretability. I am broadly interested in reducing existential risks from AI systems.
I am currently a Pivotal Fellow under Kevin Wei, conducting research into evaluation awareness and automated evaluators/auditors. Before this, I was a SPAR Fellow under Santi Aranguri at Goodfire, where I worked on interpretability methods for steering and detecting evaluation awareness, as well as understanding how training dynamics during reinforcement learning impact evaluation awareness.
Before making the career transition towards empirical safety research, I graduated from Georgia Tech and then worked on causal inference models at Meta.
I am recruiting for empirical safety roles! I am also mentoring a Fall 2026 SPAR project on Training-Time Mitigations for Eval Awareness and Eval Gaming; applications close August 18.
I also do amateur field herpetology, and love all things reptile and amphibian! In particular I am especially interested in Pythons and Anurans (frogs and toads), and believe that snakes are highly misunderstood and beautiful creatures. I enjoy learning about animal phylogenies and the evolutionary pressures that shaped them to evolve particular traits for the sake of optimizing towards reproducing (and all instrumental goals involved in reproduction, like surviving). On a related note, I find it entertaining and helpful to think about AI alignment problems in terms of evolutionary pressures and biological analogs!
A couple of links about me:
If you so desire, you can contact me via email. In general if you are considering emailing me, I would bias towards doing so! I am very accepting of people reaching out to me to chat, regardless of background. Either of these emails will reach me, but the first one is monitored more closely: