Really exciting results on paper. But truly interesting to see what data this has been trained on. There is a thin line between accuracy improvements and the data used from users. Hope the data used to train was obtained with consent from the creators