Meta’s new paper shows a new approach to language vision models

https://www.youtube.com/watch?v=hfQIuzc_WyA