it truly is shown that The straightforward pre-teaching job of predicting which caption goes with which graphic can be an economical and scalable way to understand SOTA image representations from scratch with a dataset https://k2spiceshop.com/product/liquid-k2-on-paper-online/