I’m thinking that the vision AI model, which might be specifically selected for visual analysis, is under high load. ChatGPT is working for me right now for discussing an image, but so slow and going in spurts of token production, still continuing about two words a second as I write this, and then some bursts, that I would think there’s some significant issue.
Alternately, 3AM on a holiday weekend in California is when infrasturucture is maintained and a pool of resources is partially taken offline…
(interesting, in its vision analysis that doesn’t have reluctance in identifying AI imagery as AI: ChatGPT thinks a peculiar ChatGPT image output (experimental?) is a Photoshop montage instead of AI.