honestly very excited for CLIP or something similar to be used in search engines for image search. image search currently sucks because they have to trust the surrounding text/alt text to give context
example: "dark brown box with grey lettering"
an image like this almost definitely exists on the internet, but it's unlikely to be labelled that way. it would more than likely be labelled with something like "box of hershey bars" or whatever the box *actually* contains
