I'd like to next the article which says almost no need for training models. That's really what I'm interested in at this point