pretty sure it was on micro.blog I saw an article explaining how midjourney was training on flickr creative commons images and that data set can be seen to be bias as for example there was no southern hemisphere cities in the data…. if anyone can remember this and reshare… hunting my stuff…