The only human-made images I got (2 out of 30) were also the only ones that were not square. Given that most GANs only do square images, human made ones should

The only human-made images I got (2 out of 30) were also the only ones that were not square. Given that most GANs only do square images, human made ones should be square also if we want this to be fair. And as some others have said, DALL-E 2 really seems to have a "style" that's quite easy to spot once you've seen a few images already, so even without the square hint I think would have gotten close to 30/30. But I would like to see the results on a better crafted dataset with "naive" people, like a kind of visual Turing test.