The data that they released, they uploaded a training model of Doodon GitHub, and it actually leaked personal information from the data. So I think we've seen a lot of cases where as we've been in the deep learning era, data has been king. The more and more we're seeing that data isn't so easy to bake in biases.