That’s a very interesting question. But as I’ve been using and following various forms of ‘generative AI’, I have another one. Supposing AI never becomes capable of telling the difference between fact and fiction? I’m noticing this particularly in image generating systems (but it also applies to text-generating LLM’s). The best of these, like Midjourney and Dall-e, can produce amazingly realistic images, as well as images in a whole range of artistic genres. But they also frequently include incongruent details. Yesterday I was looking at an image Midjourney had produced as a mock-watercolour painting of Central London in the rain. It was really well done — it would be hard to tell from a real painting. Except there was a red bus in the middle which had a staircase at the front. In order to know that that is not right, you have to know what a bus is, out there, in the ‘real world’. For Midjourney, though, there is no ‘out there’: it’s not painting ‘a bus’, it has no idea what a bus is. In this respect generative AI is very dream like — it creates imagery from memories, rather than perception. And, like dreams, it’s full of incongruities. Since we human beings have never been able to remove incongruities from our dreams, will AI?
Nicely put and yes, I’ve seen lots of images with people with three arms and other anomalies. As humans, even non artists are capable of spotting these incongruities and we must also remember that similar artefacts will arise in the generated prose and also code. The more worrying problem is that people most likely won’t realise in these situations and will blindly rely on the results.
Now, can someone develop an AI gas plumber as I can’t for the life of me get one to come and move a pipe.
And I thought that’s a new trend.
@Jannis regarding the initial idea in this thread, to find a way to derive colours based on a Source Grid Item position within a Poster2 layout, Stuarts initial script works well selet a colour based on the grid item number in the Poster2 List.
However, for any rid item (poster-item) that links to display the Poster2 Detail page, the position of the grid item that calls that page, doesn’t exist. So I was wondering if there is a way to access the Poster2 item number?
Eg. in a mustache tag?
Yes that might work.
And if we add both of these things together — AI using AI generated material as a source, and images of three-armed people — we end up with systems that think three arms are normal because they’ve been absorbing so many images of three-armed people… 😳
The video in this tweet made me think of this conversation lol
As soon as you’re in detail view, the information at which position the post was inside the list isn’t available any more.
The distortion of bodies is a problem that mostly is already fixed, and it can easily be fixed with negative prompts.