Kidding aside, yes, you're exactly right. We're already doing this to a large degree and getting better. Lots of our own data labeling and model training to make this good.