I mean, why don't you ask them for unfiltered models and a few billion tokens?