[MrGilbert]

This is remarkable. I'd love to see that combined with some kind of sentiment analysis like Microsoft offers, just to see if something useful comes out of it.

Sometimes, I don’t know the exact words when looking for a meme, but once I see it, I know that’s the one.

[Cloudly]

Semantic search using CLIP embeddings / other LLM embeddings could be an amazing addition too.

[astrange]

Unfortunately semantic analysis barely works at the best of times, but it especially doesn't work here. Computers… they're just not good at irony.

IME CLIP embedding search can work strangely on memes as well, because it gets confused when images have words in them. Basically the same problem reported in the original CLIP paper where it thinks an apple and a piece of paper with "apple" written on it are the same thing.