Comment by yosito

As someone who appreciates the craftsmanship and care that goes into human-created art, I found the abilities of DALL-E to be a little distressing. These limitations really point out how much DALL-E is still very different from humans, in a way that I find reassuring. Human reasoning still seems to be out of the reach of these kinds of models. What they really seem to be is very very good imitators which, after being trained on very large amounts of data, can imitate what they've seen. You could make an argument that that's essentially what humans do too, just in a more complex way. But I think it underscores why humans and machines will always be different. We'll always be trained on different inputs, and have our own unique way of combining those inputs, even as the abilities of AI rival and even surpass our own in many ways.