it truly is demonstrated that the simple pre-schooling process of predicting which caption goes with which picture can be an productive and scalable way to know SOTA picture representations from scratch on a dataset of https://k2spiceshop.com/product/liquid-k2-on-paper-online/