I also got middling results. I gave it a novel NASA picture of the moon passing in front of the earth, taken by the Deep Space Climate Observatory Satellite, so an unusual vantage point. First it was just wrong, and said that it was a picture of a crescent moon taken from a mountain top, or something like that. Not terrible at least it recognized the moon, but wrong. I tried to give it a chance and corrected it with the additional info that the picture was taken from space. It then did better, but on the last line it threw in that it was not a photograph, but a rendering of some sort, which again, is just wrong.

More and more I get the impression that people are so taken with the chat interface that they are grading these models more favorably than is warranted.