I'm curious what people think about this thesis. Personally, I'm skeptical. As amazing as DALL-E/StableDiffusion are, they require a lot of trial and error to

I'm curious what people think about this thesis. Personally, I'm skeptical. As amazing as DALL-E/StableDiffusion are, they require a lot of trial and error to generate the image you want. But this relies on you being able to look at the image and know what needs to change (and then asking for that change). The problem with programming is that you may not be able to ask for a fix without understanding the program itself. And if you have to understand the program, then the AI is just saving typing.