I feel this is correct. Google has probably the most data of any company in the world. For better or worse, they don't need to train on the outputs of other models.