I had ChatGPT generate some selfies taken in various places, then ran it through this app. My assumption was that this app would do really well, since one model

I had ChatGPT generate some selfies taken in various places, then ran it through this app. My assumption was that this app would do really well, since one model would identify the stereotypical features generated by the other model. It got 1/3. It nailed Minneapolis, it got Damascus, Syria wrong (said Amman, Jordan), and it got the Ballard neighborhood of Seattle wrong (said San Francisco).