The research group Resolution seeks to identify and control low-dimensional structures within model pretraining. This approach targets emergent misalignment and subliminal learning to prevent hidden undesirable behaviors. By systematizing persona research, the team aims to intervene in model internalizations. Practitioners can now apply for roles to help refine these alignment techniques.