[psbp]

I've been using pictures of artifacts from random museum visits through the years to test the recent vision models. GPT-V is the first that has gotten anywhere close to identifying them accurately.

It's usually able to identify 1) The materials the artifact is made out of 2) The country/region it came from 3) the significance/use of the item 4) roughly when it was created.

The images that I'm sharing are my own, so it's not pulling the images directly from the internet, and for some of the artifacts it's actually difficult to find similar images online.

I think it's fairly safe to say that it's truly able to perform advanced image analysis with images that aren't directly in its dataset.