[redox99]

I wonder how good multimodal GPT4 is at ImageNet.

(You give it the image and prompt it with the 1000 classes and ask it which one the image belongs to).

I'm surprised ClosedAI didn't include this kind of benchmark. I guess it doesn't do too well?

[sharemywin]

Here's something on Clip

https://www.pinecone.io/learn/zero-shot-image-classification...