most of the limits arw likely going to be GIGO, the same as using synthetic training data.