@wordgrammer ah hahaha i feel like it really depends where you draw the system boundaries as to whether we're really doing *less* wrt data & training- imo the data pipelines themselves & source models for synthetic could all be considered part of training......