AI crap in compo entries?
category: general [glöplog]
Good evening. Mildly interesting or relevant, but it seems ChatGPT was watching Revision compos. You probably know how its training data was (allegedly) capped in 2021. Well, he said something interesting just now, as I tested its knowledge of the demoscene.
That's some very accurate hallucination!
(For those who didn't get it: Remnants by Alcatraz was a compo entry at Revision 2024, just two days ago, and indeed in the 256 byte category, although it's not MS-DOS.)
That's some very accurate hallucination!
(For those who didn't get it: Remnants by Alcatraz was a compo entry at Revision 2024, just two days ago, and indeed in the 256 byte category, although it's not MS-DOS.)
A model trained on a million images of what is "an eye" may have taken an extremely distant view of the original "eye" of a specific artist - which could be the very vague notion of the "ellipsoid" nature of eyes in general. Or some other features in a multi-dimensional area of the problem we cannot even comprehend.
So in other words, the network could be pulling an extremely low amount of information (not even measurable in bits) from all those meticulously drawn eyes from the millions of artists. Yet manage to construct some "eye looking" image as a collage of all this.
If I were to start taking photographs in a museum and create a (horribly looking) collage of these photographs, where each one takes (lets say) 5x5 almost unrecognizable pixels (because it's too small or blurred or whatever), would it be the same ?
So in other words, the network could be pulling an extremely low amount of information (not even measurable in bits) from all those meticulously drawn eyes from the millions of artists. Yet manage to construct some "eye looking" image as a collage of all this.
If I were to start taking photographs in a museum and create a (horribly looking) collage of these photographs, where each one takes (lets say) 5x5 almost unrecognizable pixels (because it's too small or blurred or whatever), would it be the same ?
i'm sure you'll eventually find the one carefully constructed tortured theoretical example that will make all the crummy stuff happening thousand- and millionfold already okay to do 👍
Quote:
btw on the matter of "it will drive pixellers out of job", I've heard this before for "true" coders: DOS vs windows, rasterizers vs Opengl, C++ vs Unity/unreal/notch.
It will inevitably shift the profile of the artist. Further away from the classicaly trained visual artist into prompt jockeying SEO optimizing trained engineer. Away from the crafty hands-on DIY type into data analyst.
For example, You know all that nice 4D footage that was used in your winning The Legend of Sysiphus demo? This AI tech aims to throw out exactly the people that actually did the gig (You?). Starting with the actor and not stopping all the way to the cable rolling guy. Deskill them all and hire prompt jockeys. For all AI companies care these creative people can either become prompt jockeys themselves, or just deliver food to them. That’s sad isn’t it?
Quote:
i'm sure you'll eventually find the one carefully constructed tortured theoretical example that will make all the crummy stuff happening thousand- and millionfold already okay to do 👍
There may be other crummy stuff happening with copyrights and whatnot, but the truth of the matter is still that the NN does not copy specific pixels from a specific set of images, but some deeper understanding of the image in the way we cannot comprehend. Therefore there is a giant gray area for using these things in the creative process, that doesn't just involve "Here is a scrolling image" but lots more (* for example, a technique to tile specific textures).
A solo coder prod using "AI" art and a random music track off the internet, is perfectly valid to me, as is a solo graphics artist prod playing back animated 3D scenes with unity.
The second one is not impressive or interesting to me as a programmer, as I'm sure the first one would not be impressive or interesting to a graphics artist, but that's all up to each individual to judge and vote.
I downvote cables/notch/unity prods, you can downvote prods with AI art, but all this butlerian jihad against AI is taking it a bit too far, and the "took-our-jobs" mentality is laugable.
The second one is not impressive or interesting to me as a programmer, as I'm sure the first one would not be impressive or interesting to a graphics artist, but that's all up to each individual to judge and vote.
I downvote cables/notch/unity prods, you can downvote prods with AI art, but all this butlerian jihad against AI is taking it a bit too far, and the "took-our-jobs" mentality is laugable.
Quote:
It will inevitably shift the profile of the artist. Further away from the classicaly trained visual artist into prompt jockeying SEO optimizing trained engineer. Away from the crafty hands-on DIY type into data analyst.
For example, You know all that nice 4D footage that was used in your winning The Legend of Sysiphus demo? This AI tech aims to throw out exactly the people that actually did the gig (You?). Starting with the actor and not stopping all the way to the cable rolling guy. Deskill them all and hire prompt jockeys. For all AI companies care these creative people can either become prompt jockeys themselves, or just deliver food to them. That’s sad isn’t it?
I'm sorry and with all my respect - I still don't see how this is different than the old luddite
arguments (tracked vs mp3, dos vs opengl, code vs noodling in notch): "But it will be the death of xyz component of a demo group". or "It will be too easy to make a typical demo" .
Maybe two advices - Adapt or perish. And let others be.
The stronger argument in this conversation is the original from Gloom - which is to say, under certain rules etc etc for legal reasons at a party bla blah. This is a practical issue that I can understand.
Quote:
Adapt or perish.
Silly me for thinking we’re past social darwinism. Especially in a niche hobby.
Quote:
The giants on whose shoulders the generators stand are not acknowledged (or remunerated) as giants, but instead must now compete against perhaps for the moment slightly faulty but cheaper and more productive ghosts of themselves, built from the shards of broken mirrors.
Please point me to the good part of this truth.
The "3D flyby of the 2020s" aspect also came to my mind. It's still different though, since we could smell those old 3D flybys coming from a mile away.
Quote:
I'm sorry and with all my respect - I still don't see how this is different than the old luddite arguments (tracked vs mp3, dos vs opengl, code vs noodling in notch): "But it will be the death of xyz component of a demo group". or "It will be too easy to make a typical demo".
I guess the difference is that in those cases humans were still creating things, in different ways but it was all made by people. with AI art, the creation is diminished to 'writing some good words into an LLM' and then an image is created with stolen data. Even making a demo in a pre-existing engine needs more and understanding of the craft than that.
Also, some people have always approached the scene as very code heavy, only programming being a truly noble thing, art and music being optional. From that aspect yeah, go wild with your generated garbage, but for other people, it's the whole package, music and art have just as much merit as programming, and then it's just a fucking eyesore to have to watch contrived AI bullshit on screen.
In the end people are free to do what they want and everonye is free to call out if it sucks, as is happening with that amiga comic thing.
I think the creative use of AI (tools) should be honoured in one way or another as a means of expression, if there is other own created content (code, music, effects) to back it, just as well as actual skills should be honoured separately.
Maybe it's time for a new set of compos and rules to differentiate between creativity and skill (there may be overlaps, mind you). More importantly however: people who enter material into a competition should be honest about their approach. If a 2 hour prompting session wins over a 2 week of sweat-and-tears shed hand pixeled graphics goodness, the whole competitive aspect of a demoparty compo gets lost in an instant.
Maybe it's time for a new set of compos and rules to differentiate between creativity and skill (there may be overlaps, mind you). More importantly however: people who enter material into a competition should be honest about their approach. If a 2 hour prompting session wins over a 2 week of sweat-and-tears shed hand pixeled graphics goodness, the whole competitive aspect of a demoparty compo gets lost in an instant.
Quote:
Adapt or perish. And let others be.
Things / people who perish tend to let others be, I'll give you that much.
As much as I despise quote dropping, since someone mentioned Butlerian Jihad, I just have to do it: "The target of the Jihad was a machine-attitude as much as the machines," Leto said. "Humans had set those machines to usurp our sense of beauty, our necessary selfdom out of which we make living judgments."
This seems like a reasonable way forward to me.
Quote:
Maybe it's time for a new set of compos and rules to differentiate between creativity and skill (there may be overlaps, mind you). More importantly however: people who enter material into a competition should be honest about their approach. If a 2 hour prompting session wins over a 2 week of sweat-and-tears shed hand pixeled graphics goodness, the whole competitive aspect of a demoparty compo gets lost in an instant.
This seems like a reasonable way forward to me.
Quote:
somewhere before the creation of the model trained on art without the informed, enthusiastic consent of the artists
So AI trained with material freely provided by artists is okay?
Quote:
Quote:Adapt or perish.
Silly me for thinking we’re past social darwinism. Especially in a niche hobby.
Quote:The giants on whose shoulders the generators stand are not acknowledged (or remunerated) as giants, but instead must now compete against perhaps for the moment slightly faulty but cheaper and more productive ghosts of themselves, built from the shards of broken mirrors.
Please point me to the good part of this truth.
Adapt or perish - ok, not perish literally ! one can still use DOS and stuff in 2024 and that's fine, as long as there is no crusade against those who do not. These are all affairs in development and it's good that we talk about it. Lets see what happens, I'm neither against or for it personally but I'm definitely against calling anyone a lamer because they use X or Y, or forcing anyone to do or not do something in a specific way - the technical legality of the issue is a different thing.
The last part of the quote - I don't know what to say. This has happened many, many times before in history. I don't know if it is a good or a bad thing, or a necessity of capitalism or whatever else. It is certainly a good thing for some (those who work on this field), not so good for some others (the artists). Maybe I'll have an opinion in 5 years from now when all is more developed and settled. Maybe we will have a myriad of "AI inspired" demos from now one and our judgement and thumb up/down ratios will develop accordingly. Just like with a demo made with Notch - some will say " Hey I work for Notch, this is my code" to justify an otherwise not so pleasant attack from "some" who will say "you lamers this is just point and click some nodes, 100% of the work made by others".
This was more evident a few years ago, I think it has settled quite a bit now.
Quote:
Adapt or perish.
i've typed and deleted four answers to this. time to take a breather and bow out.
Ask yourself what do you want?
Do you want to look at art made by a human hand, pixel by pixel. Probably full of soul and the artist spend a lot of time on it and had their reason for making all those choices in the image. Also reaching this level of art has probably taken a lifetime.
Or do we want to look at a soulless, meaningless quick prompted AI image done with no skills at all?
If you think AI is fine and if people accept it, almost all human artists will quit the scene.
Do you want to look at art made by a human hand, pixel by pixel. Probably full of soul and the artist spend a lot of time on it and had their reason for making all those choices in the image. Also reaching this level of art has probably taken a lifetime.
Or do we want to look at a soulless, meaningless quick prompted AI image done with no skills at all?
If you think AI is fine and if people accept it, almost all human artists will quit the scene.
Quote:
It is certainly a good thing for some (those who work on this field)
In this case I think it’s good only for the investors/shareholders. But only they matter right? Not forum jerkoffs like us here. I’m not trying to fight the tide here, I’m just saying that I feel tide=bad. In this case. Please don’t go philosophical with “but tide means life for some creatures”.
Quote:
If you think AI is fine and if people accept it, almost all human artists will quit the scene.
Reading that from a respected hard-working long-time graphics artist makes me feel sad and it shows me where this can lead to if we really accept it.
Quote:
Quote:Adapt or perish.
i've typed and deleted four answers to this. time to take a breather and bow out.
and you did the right thing. I didn't want to write perish. It's an expression that goes with "adapt" and I got carried away on the keyboard. I apologise if I offended somebody, it wasn't may intention. "perish" doesn't make any sense in the scene. Old hardware and old ways is still around, and long may be. Adapt or not if you want, that's ok. But give others the space to decide for themselves if they also want to.
Quote:
Quote:It is certainly a good thing for some (those who work on this field)
In this case I think it’s good only for the investors/shareholders. But only they matter right? Not forum jerkoffs like us here. I’m not trying to fight the tide here, I’m just saying that I feel tide=bad. In this case. Please don’t go philosophical with “but tide means life for some creatures”.
There are also many, many engineers who make a living out of all this (* not restricted to LLM image generation) and many more who benefit from all the advancements. You mentioned our demo "sisyphus". I can tell you that AI was been used for it (licensed, I can give specific details) to do full body motion tracking and hence improve massively the compression ratio ( without which the demo would be many times larger still). So in a sense without "AI" (broadly speaking) there would be no demo, or a very different demo that was not my vision or whatnot. There will be more and more clever uses of AI; going back to my original question - is a "tiling" or "clever inpainting" use of AI really much different to the amiga demo ?- but, after all, the inpainting WAS trained on "other people's dataset". This is undeniable. Yet, we will turn the blind eye; or not!? Where do you draw the line ?
Quote:
Or do we want to look at a soulless, meaningless quick prompted AI image done with no skills at all?
Don't know why AI-art seems soulless to you. When I don't know if a picture was drawn by a human or an AI, it's the same picture. If I take a photo of the Mona Lisa it's still the same motive. It doesn't get less good because I made a reproduction. (apart from the fact that i cannot see the brush prints.
would not believe anything that what gpt says, that thing just tries to be nice and has hallucinations.
AI can just blend existing stuff but not come up with new stuff.
its probably a coincidence do the same in a new chat in get a different result
It will do anything for an applause. If there are no facts about something it will dream them up.
With all the respect in the world to all the different people here. Could we please keep this discussion within the demoscene boundaries? Stretching it to "all art is doomed" real-estate isn't very productive. If you absolutely feel the need to stretch it, then please at least keep it within the digital content realm which is the field of AI generation anyway. Art (with capital 'A') is a different league, a different discussion.
