From what i understand they don’t have to worry about IP so their training data can be much larger