Now

Hello, I'm David Crispell.

I am a safety researcher interested in generalization and agent modeling, currently investigating subliminal learning.

I am especially interested in how reinforcement learning aligns model behavior towards goals, and how those mechanisms matter for scalable oversight.