I’m Thomas Jiralerspong, an AI safety and mechanistic interpretability researcher. I write about research, building with AI, and life as a researcher.
Subscribe to get new essays by email.
Three places to start
Chasing the Karpathy Peak — How I became a slopaholic, who’s most at risk, and how to avoid it or recover.
Persona Vectors don’t work on real data — Why evaluations on synthetic data can miss what happens in real conversations.
The Art of AI Art — On creative vision, creative execution, and the art I haven’t made yet.
More to explore
My personal website has my research, projects, and mentorship work. Earlier website versions of my essays remain in the blog archive.