It will be interesting to watch how they decide what new data to train on if most of it is low quality.