pouët.net

Exactly. Problem I see is that content gradually forms expectations. So, yes, LLMs will be used for generating mediocre/half-retarded/worthless sh*t. But the capacity of LLMs for sheer volume of such sh*t plus the speed of production threatens to flood everything else out and form a new generation of halfwits. I mean nobody was into Big Brother shows when they started, right? I think (more or less) the only thing one can really be sure of is that LLMs are going to furtherly debilitate the average earthling.

I just read an article at forskning.no (in Norwegian) that says people can't distinguish AI art from human created art. However when told the art was AI generated they perceived it as uglier than if they were told a human created it. Seems people like that there is a human behind the art and that the art has a backstory of some kind.

When making something for the demoscene I also feel the same way. On our J'_ demo we had a bunch of pictures of pipeworks, circuit boxes and hallways. If they were taken off the internet I would feel nothing about them, but since the photos were taken at my university venturing into some "off limits" area where we also found the janitor's porn magazine cache these pipes suddenly means something more :D

Here is the paper in English
https://www.nature.com/articles/s41598-024-54294-4
and the norwegian article which you will not read.
https://www.forskning.no/kunst-kunstig-intelligens/folk-likte-ki-kunst-nar-de-trodde-den-var-laget-av-mennesker/2330420

Many think that one of the essential properties of the art is that it is communication between humans. Even before AI you weren't always able to separate a piece of art from a randomly generated material but if you knew it's the former then you could reflect what the creator is trying to communicate (in some more concrete or an abstract aesthetic sense).

(now using AI as a tool ofc doesn't dispose this communication dimension entirely)

hAIrpsi lead to no result?
Knowing the rules.

It's all nice if you are a pro, but what if you cannot do better than AI :P

then you keep training/trying and respect those who got really good at it.

Falls right into my personal definition of “art” I discussed here before. Art needs context. Without context it’s a mere decor.

Prompt gives context.

Sure man. “Beautiful flower in style of Van Gogh” or “cute cat with big eyes comic book style” provides context to the cr*p you get out of the LLM like there’s no tomorrow. And that makes you an artist for sure. I thought context required, you know, understanding…

I just adore seeing something like this at linkedin:
- data analyst
- software engineer
- backend developer

…and newly added:
- AI artist

I don’t know whether to laugh my ass off or cry.

Latest LLMs can use up to 1 million tokens in a prompt. So you could actually provide very elaborate context. Maybe it doesn't work yet so well in current gen image/video generators, but it's coming.

“understanding”

One question: Is the prompt writer the artist or the machine that “paints” the picture?

The "artist" today in a broad meaning is a *person* that create things (and many more things people consider artists to be). In this sense, neither prompt engineer nor machine is an artist. It's a new world, where a credit for artistic creation is shared credit between prompt engineer, AI model developers, and obviously artists who contributed to training data. Maybe, if you stretch your perception of AI, the AI model "itself" can also be considered an author, but AI models so far has no state/entity/awareness etc...so I would say, so far, it's just a tool.

So prompt engineer is more like an artistic director really.

I've been pumpin' like hell to get more power to my hand, so the AI cannot force it.

I agree that the current concept of “art authorship” is desperately outdated. This has been discussed in the art circles for decades. And it’s an ongoing discussion. However, the people behind this wave of “prompt-to-image” toys were never part of that discussion. In fact, I doubt they are even aware of any philosophical discussion on concepts of art and authorship. I don’t think they are versed in writings of Benjamin or McLuhan. Really, they have no place in this discussion, so it’s insulting to suggest we redefine the concept of autorship according to what suits the tech bro’s newest cash cow. Big gamechanging concepts can’t come a posteriori to tech releases.

Once again, whether it's art or not is the least of the problem.

whether something is art (or not) is not up the artist but his or her audience.

+to

but @tomkh: good summary and agreed. it's yet another tool that requires human supervision.

..and a Chrono Trigger quote since it seems so fitting:

Memories of Green

I would differentiate between popular art and more ambitious art sometimes called high-art.

According to my ad-hoc definition, popular art is artistic content deliberately created to reach the largest audience possible, while high-art is more about pushing the boundaries of creation with the usual audience assumed to be educated in art history and sophisticated enough to understand the depth of the creation, context etc...

This doesn't really mean any of the kind is higher or lower quality.

In fact, popular art has pretty high quality bar, because quality helps to reach widest audience possible. However, it is not as demanding when it comes to originality. For example, it's quite common that very generic song gets to the top charts,because the singer has already a strong brand. This is where AI could potentially shine (if done right). It could create high-quality, but pretty generic content. Maybe useful in gamedev too to cut cost.

For high-art, originality is much more important, and AI struggles with it, even more than one would expect from such systems. Extensive prompt engineering won't help you that much, so yeah, I was stretching it a little bit in the previous posts. But who knows, it may improve in the future.