Yeah, I got one. It does have a lot of trouble with them, which isn't very surprising, given snakes are not something it is particularly good at in general.
Ouroboros has issues for the same reason why gryphons often had issues - gryphons are cats mixed with birds, but most images of cats and birds are, you know, cats and birds, not stran…
Yeah, I got one. It does have a lot of trouble with them, which isn't very surprising, given snakes are not something it is particularly good at in general.
Ouroboros has issues for the same reason why gryphons often had issues - gryphons are cats mixed with birds, but most images of cats and birds are, you know, cats and birds, not strange fantasy hybrid creatures. As a result, it can have issues with trying to turn the gryphon into a cat or a bird in its entirety rather than creating the proper mythological creature, because a cat or bird is more "coherent".
The same applies to the Ouroboros - a snake eating its own tail is something that "confuses" it because you're trying to tell it to do something that's "wrong" (which violates usual coherency rules). You can get it to do it, if you know how, but it takes more effort than it does to produce more "normal" images.
The AIs have gotten vastly vastly better at this sort of thing over time. For instance, V3 couldn't produce an adult dragon at all (it could sort of produce baby dragons, but they were pretty bad); by V5 it was producing pretty good adult dragons and V6 is even better at it. Gryphons have gone from "nearly impossible to produce a full body shot of" to "pretty easy to produce a full body shot of".
And you can definitely produce things that aren't in the data set, like producing 40K style miniatures of things that are wildly not 40K (like for instance, alicorns). It's fun!
Your notion that it doesn't understand "concepts" is flawed. It doesn't really "understand" anything, but "concepts" aren't actually any harder for it than anything else to produce images of as long as the "concept" has some sort of consistent visual representation.
You thought your "no elephants" prompt was clever but I rather easily produced an empty room with no elephants by simply telling it not to include elephants, and then you tried to move the goalposts rather than admit you were wrong.
I would expect that many of these AIs will be able to produce something like an ouroboros pretty reliably in a year or so, as they've already gone a very long way in terms of producing a lot of other weird mythological things. But it wouldn't mean they're "intelligent" in any way, because they're not. That doesn't mean they're not useful.
Incidentally, another problem with the ouroboros is that it is not actually drawn particularly consistently - sometimes it is drawn in a single circle and sometimes it is drawn in an infinity sign, or some other configuration. If you specify more specifically what you're looking for, you get something a lot closer - for instance, when I did "Digital painting of an ouroboros snake eating its own tail" while giving it a visual reference of what I was looking for and a proper aspect ratio for the sort of image I was looking for, I got much better results.
The standard idea - an abstract idea, a general notion - is absolutely something that AIs deal with. They don't actually "understand" them, but a language-based prompt is by definition pretty much made up of concepts. Like if you tell it to draw you an anthropomorphic parrot pirate, that's a bunch of concepts.
If you had literally typed concept definition into Google, you would have immediately seen the definition I used. Saying it is not the "actual definition" is quite silly.
Secondly...
Styles are a form of concept - a style like "watercolor painting", "In the style of anime", the style of a particular artist, a cartoon, a pixar film, Disney, My Little Pony, Warhammer 40,000, etc.
There is a particular concept of what something in a given style "should" look like.
AI art programs are capable of producing things in particular styles. Ergo, they can do "concepts".
Indeed, you can even mix these things together and end up with a My Little Pony-looking unicorn wearing Warhammer 40,000-style space marine power armor.
Another "concept" would be "anthropomorphic", a word I use all the time to generate anthropomorphic animals.
As such, art AIs, at least, "understand" the concept of style, and are capable of applying it to images to produce images in particular styles. Though of course, they aren't actually intelligent, they are indeed capable of working with concepts in order to produce images that match the prompt.
Yeah, I got one. It does have a lot of trouble with them, which isn't very surprising, given snakes are not something it is particularly good at in general.
Ouroboros has issues for the same reason why gryphons often had issues - gryphons are cats mixed with birds, but most images of cats and birds are, you know, cats and birds, not strange fantasy hybrid creatures. As a result, it can have issues with trying to turn the gryphon into a cat or a bird in its entirety rather than creating the proper mythological creature, because a cat or bird is more "coherent".
The same applies to the Ouroboros - a snake eating its own tail is something that "confuses" it because you're trying to tell it to do something that's "wrong" (which violates usual coherency rules). You can get it to do it, if you know how, but it takes more effort than it does to produce more "normal" images.
The AIs have gotten vastly vastly better at this sort of thing over time. For instance, V3 couldn't produce an adult dragon at all (it could sort of produce baby dragons, but they were pretty bad); by V5 it was producing pretty good adult dragons and V6 is even better at it. Gryphons have gone from "nearly impossible to produce a full body shot of" to "pretty easy to produce a full body shot of".
And you can definitely produce things that aren't in the data set, like producing 40K style miniatures of things that are wildly not 40K (like for instance, alicorns). It's fun!
Your notion that it doesn't understand "concepts" is flawed. It doesn't really "understand" anything, but "concepts" aren't actually any harder for it than anything else to produce images of as long as the "concept" has some sort of consistent visual representation.
You thought your "no elephants" prompt was clever but I rather easily produced an empty room with no elephants by simply telling it not to include elephants, and then you tried to move the goalposts rather than admit you were wrong.
I would expect that many of these AIs will be able to produce something like an ouroboros pretty reliably in a year or so, as they've already gone a very long way in terms of producing a lot of other weird mythological things. But it wouldn't mean they're "intelligent" in any way, because they're not. That doesn't mean they're not useful.
Incidentally, another problem with the ouroboros is that it is not actually drawn particularly consistently - sometimes it is drawn in a single circle and sometimes it is drawn in an infinity sign, or some other configuration. If you specify more specifically what you're looking for, you get something a lot closer - for instance, when I did "Digital painting of an ouroboros snake eating its own tail" while giving it a visual reference of what I was looking for and a proper aspect ratio for the sort of image I was looking for, I got much better results.
You didn't "tell it" to do anything. You used a software switch.
None of the above changes my position one iota- AI don't deal with concepts. The reason why was deal with by Searle decades ago.
What definition of concept are you using here?
The standard idea - an abstract idea, a general notion - is absolutely something that AIs deal with. They don't actually "understand" them, but a language-based prompt is by definition pretty much made up of concepts. Like if you tell it to draw you an anthropomorphic parrot pirate, that's a bunch of concepts.
No. Look at the actual definition. There's nothings specific that's generalized from particulars https://www.merriam-webster.com/dictionary/concept and no again- A prompt to an LLM is simply a pattern of characters. Please familiarize yourself with how those things actually work. https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/
If you had literally typed concept definition into Google, you would have immediately seen the definition I used. Saying it is not the "actual definition" is quite silly.
Secondly...
Styles are a form of concept - a style like "watercolor painting", "In the style of anime", the style of a particular artist, a cartoon, a pixar film, Disney, My Little Pony, Warhammer 40,000, etc.
There is a particular concept of what something in a given style "should" look like.
AI art programs are capable of producing things in particular styles. Ergo, they can do "concepts".
Indeed, you can even mix these things together and end up with a My Little Pony-looking unicorn wearing Warhammer 40,000-style space marine power armor.
Another "concept" would be "anthropomorphic", a word I use all the time to generate anthropomorphic animals.
As such, art AIs, at least, "understand" the concept of style, and are capable of applying it to images to produce images in particular styles. Though of course, they aren't actually intelligent, they are indeed capable of working with concepts in order to produce images that match the prompt.
No, it's not silly. It's the Merriam Webster dictionary.
I've already shown you the article explaining how it works. If you don't accept it then so be it.