Now

Hello, I'm David.

I am an alignment researcher interested in generalization and agent modeling, currently investigating subliminal learning.

I am especially interested in how training dynamics shape model behavior, and how those mechanisms matter for scalable oversight.