AI tooling in the Demoscene
category: residue [glöplog]
If they make something really cool using AI, all power to them IMHO
So far i have only seen boring, mostly mediocre at best, though
So far i have only seen boring, mostly mediocre at best, though
> i totally get that it's tempting to get rid of a tedious task that just seems annoying and "has to be done".
I agree that there is such a thing as being too lazy, and I think that is actually a large underlying part of the discussion. We're interested in human effort and the change that brings in humans. That's part of the art.
However... I don't really think your argument makes sense, because it can be applied just as well to Unreal Engine, Crinkler and any tools you use to paint. They are all just tools we use because they save us time. And we accept those tools because the leverage they create does not shortcurcuit the thing about the art that we are interested in.
Whether the tool is written by AI or not is irrelevant. The thing that matters is whether its leverage goes in a direction where the art becomes uninteresting.
For example use of Unity and Unreal Engine, which there is mixed opinions about. They can make the art uninteresting because "Well what did you really do?" If it's not clear that the author really did anything but just used builtin stuff in Unreal to create something pretty - well then the product will rightfully judged to be poor.
I agree that there is such a thing as being too lazy, and I think that is actually a large underlying part of the discussion. We're interested in human effort and the change that brings in humans. That's part of the art.
However... I don't really think your argument makes sense, because it can be applied just as well to Unreal Engine, Crinkler and any tools you use to paint. They are all just tools we use because they save us time. And we accept those tools because the leverage they create does not shortcurcuit the thing about the art that we are interested in.
Whether the tool is written by AI or not is irrelevant. The thing that matters is whether its leverage goes in a direction where the art becomes uninteresting.
For example use of Unity and Unreal Engine, which there is mixed opinions about. They can make the art uninteresting because "Well what did you really do?" If it's not clear that the author really did anything but just used builtin stuff in Unreal to create something pretty - well then the product will rightfully judged to be poor.
Quote:
if it's not clear that the author really did anything but just used builtin stuff in Unreal to create something pretty - well then the product will rightfully judged to be poor.
My take is if AI is used the outcome should be simply judged more harshly. Use AI, disclose, but know that this raises expectations. Exact same thing should apply for UE demos - if I ever see this default mannequin in a demo, instant down-vote.
However, it's just stupid if ppl down-vote just for the very fact AI or UE was involved.
Quote:
You do realize that if the internet says this is valuable art, then the LLM will claim the same? You do realize that the LLM image generator is trained on this? In other words, any "problem" you ascribe to "humans and art" transfers to (and even amplifies in) "AI and art".
Yes, I do realize that. And your point is? Does it mean I have to agree? No.
There has been a lot of romanticizing in this thread about the "sanctity" of human craft. My point is that the human baseline you are defending is already perfectly capable of producing, elevating, and buying into absolute nonsense. We all know AI is trained on human biases and flaws. If an LLM amplifies that, it is just a mirror reflecting the underlying human disease. If the human art establishment is full of pretension, the AI will be too. You're basically saying, "AI is trained to be just as absurd as the humans you're criticizing". I agree, but that doesn't excuse the humans who started the joke.
You're trying to use my critique of human art critics as a weapon against AI, but I'm just pointing out that the "human touch" isn't inherently immune to being complete bullshit.
Yeah, well, sorry, when you wrote “the problems with humans and art” I didn’t see the point in saying that if it doesn’t imply “as opposed to AI and art”. So I went that way.
Anyway, as you rightly said here, we should be really talking about “craft”. Art doesn’t necessarily include good craft. AI seeks to get rid of craft. But craft is almost all we have here in this subculture. We rarely dabble in art. In my book, both art and craft imply human experience. Experience of a real human individual, not the average of all human made art. Take that away and what you’re left with is neither craft nor art. But we’re going offtopic again.
Anyway, as you rightly said here, we should be really talking about “craft”. Art doesn’t necessarily include good craft. AI seeks to get rid of craft. But craft is almost all we have here in this subculture. We rarely dabble in art. In my book, both art and craft imply human experience. Experience of a real human individual, not the average of all human made art. Take that away and what you’re left with is neither craft nor art. But we’re going offtopic again.
Tooling. Is that something old people do nowadays? There are plenty of engines that allow excellents visuals and scripting. Get with it Granddad. Engines aren't going anywhere, so everyone should just accept Unity demos instead of burying your head in the sand making tools like some old person trying to prove themself.
;)
History of cheating:
1. The Demoscene came about from users running demos just to see them. The nice demos got spread and their authors and their group built a reputation.
2. Perhaps inevitably, competitions came about to "really prove who's the best", when instead it causes frustration among those who are not and can't be the best.
3. Thus, inevitably, there were cheaters. The evidence is overwhelming. A counter-movement started almost immediately and organizers and artists alike were greatly upset at cheating in art (copy, etc) and music (covers, long samples). Extremely few know performance limits of platforms, so cheating in code (copying code, presenting animation as effect, lines.bin, etc.) is still awaiting this counter-movement.
4. Organizers make the rules and are responsible for preselecting all violators of the rules (cheaters).
5. I suggest separate AI compos, but there will always be cheaters, as in this year's Revision. Another alternative is to have the same competitions, but to now allow generated content to compete with human creations.
6. Cheating was always available, and is available now. Just take another artist's work and say you did it. For a Fiverr you can with big prizes and a Meteoriks award, if you can find the actual talent (instead of you).
7. All cheaters know what they did.
Also, I mean nobody is forcing you to be a Demoscener. You could go into the commercial zektor. They don't care where their stuff comes from, or that they can't Copyright their products anymore if they use AI.
If you're a Demoscener or an artist though, I recommend to not kill all rights to your work by using AI and thereby losing your IP.
;)
History of cheating:
1. The Demoscene came about from users running demos just to see them. The nice demos got spread and their authors and their group built a reputation.
2. Perhaps inevitably, competitions came about to "really prove who's the best", when instead it causes frustration among those who are not and can't be the best.
3. Thus, inevitably, there were cheaters. The evidence is overwhelming. A counter-movement started almost immediately and organizers and artists alike were greatly upset at cheating in art (copy, etc) and music (covers, long samples). Extremely few know performance limits of platforms, so cheating in code (copying code, presenting animation as effect, lines.bin, etc.) is still awaiting this counter-movement.
4. Organizers make the rules and are responsible for preselecting all violators of the rules (cheaters).
5. I suggest separate AI compos, but there will always be cheaters, as in this year's Revision. Another alternative is to have the same competitions, but to now allow generated content to compete with human creations.
6. Cheating was always available, and is available now. Just take another artist's work and say you did it. For a Fiverr you can with big prizes and a Meteoriks award, if you can find the actual talent (instead of you).
7. All cheaters know what they did.
Also, I mean nobody is forcing you to be a Demoscener. You could go into the commercial zektor. They don't care where their stuff comes from, or that they can't Copyright their products anymore if they use AI.
If you're a Demoscener or an artist though, I recommend to not kill all rights to your work by using AI and thereby losing your IP.
Quote:
I suggest separate AI compos
And once you have one (which already happened) please no stupid backlash. Like last time there was surprised pickachu face that on AI-allowed compo there were full AI entries (Wild with AI on ASM Winter'06). Here is example of evilpaul in denial, and here is darya claiming an AI prod posted on AI-dedicated compo is cry for attention. No wonder others would rather hide the fact they have used any AI.
Quote:
there will always be cheaters, as in this year's Revision
Exactly. That's why I'm all for safe space here. And if compo allows AI it's not cheating BTW
BoyC entry was not violating any rules, so big kudos to him to disclose some help from AI to complete the prod on time. It's just sad in today's climate one have to rather conceal the fact to avoid "AI advertiser" label.
Personally I do not encourage anyone to (ab)use AI for coding and as I said multiple times, I prefer not to use it at all for my personal projects - simply because I like to code and want to stay sharp. But people will be using it no matter what, for both tooling and actual demo, and there is no way to verify. So the worst thing we can do is lie about it.
I suggest separate demoscene websites for AI discussions too
Lol fuck AI.
Quote:
If only there was some kind of "hide residue topics" option to tick in something like a "custöomizer" or similar, to make the problem go out of sight.I suggest separate demoscene websites for AI discussions too
no, residue topics aren't the problem
Maali: what is the problem? Maybe there is no problem!
tsk, is that the best your buddy chatgpt could come up with?
Maali: I don't want to insinuate anything, but since you very much do...it seems to me you would even like to experiment with AI, but since you (correctly) recognized it's not the smartest thing to do politically in today's demoscene climate, you eagerly took the side of "the loudest voices in the room" (not to be confused with majority opinion) which are opposing AI in any shape or form.
And apart from your snarky comments and jokes you do around it, I'm actually curious what is honest opinion about the AI usage in the demoscene context topic.
And apart from your snarky comments and jokes you do around it, I'm actually curious what is honest opinion about the AI usage in the demoscene context topic.
*what is your honest opinion (made a typo)
nice try but my loud voice is more regarding how superboring and tone deaf you are with your constant AI whiteknightery. I am not against AI per se, just you and your "discussion" attitude. so yeah, now you can cry again that it's ad hominem and that the demoscene illuminati are trying to silence you. fuck you and lol, fuck AI. :D
Maali: I recently mainly throw AI as a sort of running joke. And it's not that anyone is trying to silence me, but instead of debating, the argument is exactly this what you said, that I "keep brining it up". And in fairness I'm not even initiating those discussions. But yes, I'm interested in the topic, and I did wrote a lot about it, mostly because I'm one of the few more outspoken people here who has unpopular opinion, so it naturally goes back and forth.
Another reason is essentially you are constantly teasing me - and we both know you like to do it and you do it intentionally.
Another reason is essentially you are constantly teasing me - and we both know you like to do it and you do it intentionally.
well, you can stop now! thanks!
Quote:
as of today, proper tools for intros/demos cannot be oneshot with AI, and it will not for quite a while.
I experimented a bit with fully AI generated demos for an absolute niche platform, the Raspberry Pico RP2350. This is a very powerful microcontroller where you basically have to bitbang all IO. There are a few demos for this platform on pouet.
This is basically a one-shot demo where I gave no instructions other than "make cool demo"
https://github.com/cpldcpu/PicoDemos/tree/master/11_FlashDemo
Some more examples here:
https://github.com/cpldcpu/PicoDemos
I believe this worked very well, because I primed the work space with other demos first. Gemini 3.5 Flash may also have some secret sauce from Googles world models (Omni) Subsequental demos got rather derivative, so not that interesting.
No worries, I am not planning to add this to pouet. But I guess it will be rather diffuclt to keep AI out of everything. Especially the learning curve for new and obscure platforms is basically removed. Wild compos may be an early victim here - or stick to physical entries.
As said somewhere else: Maybe the demoscene is not about limitations anymore...
What does it mean to one-shot anything? It's depressing. First of all I don't believe it. Others are struggling to make a tiny code portion of a whole project work and they babysit the LLM, so how did you or anyone claimed to one-shoot? Is it just social media engagement? Whenever I hear some post of "This is Amazing, yesterday I woke up and told chat-GPT to do something while making a sandwich. And when I came back it finished it properly" I am like Cool Story bro - Press X to doubt.
But suppose it is like this. What's the point. Click button to generate thing? What's left to the human.
Why do people want to "claim one-shot" and talk about models like they are brands all over the place? Is the internet fake and gay?
But suppose it is like this. What's the point. Click button to generate thing? What's left to the human.
Why do people want to "claim one-shot" and talk about models like they are brands all over the place? Is the internet fake and gay?
Quote:
I believe this worked very well
I like how the scroller part ends just as the greetings begin. Very prescient.
Quote:
First of all I don't believe it. Others are struggling to make a tiny code portion of a whole project work and they babysit the LLM, so how did you or anyone claimed to one-shoot? I
A response:
https://github.com/cpldcpu/PicoDemos/tree/NewDemo/12_DIRTY_MINDSET
I stored prompts here:
https://github.com/cpldcpu/PicoDemos/tree/NewDemo/12_DIRTY_MINDSET/prompts
This was using Antigravity as an agentic code hardness, others may work even better. One key element is to provide examples for in-context learning, which were there in form of the other demos. I could explain more, but maybe rather not... (see below)
(lmk if you want me to delete it)
Quote:
What does it mean to one-shot anything? It's depressing.
Yes, it seems to bring the same dread that image generators brought to visual artists to coders. Its fascinating to play with, but it completely removes most of the reasons that make coding demos compelling.
I don't deny one can orchestrate these things to do stuff.
But I wonder, how long did it take? Does the agent work on things and self checks if everything is correct, like going in a feedback loop till it gets it right? How long does that take? Are there some unit tests or something?
Of course if it finished I see you ask it to change things and then it goes again.
Also,. if I understand, there are a lot of examples of previous effects which code might it be reusing. So, it's not like from a zero state it wrote algorithms for a lot of these effects without some previous examples? It's looking at the previous codes and grabs things and scripts them together, maybe change parameters?
The question is. Because this one-shoot can be misleading, people will think "If I go now on ChatGPT or even the more professional code LLMs, and I say "Please make a cool demo"" or even give some more thorough description as a user, like "make demo with this and that effect and that music and that everything",. from what I know it's not gonna give you that. You are for pain.
But maybe one has to build sophisticated system, with previous prompts and previous examples of other demos and effects, have it base it in that codebase to build a variation of it? Some of the effects in this seems to be variation from previous (lines, mandelbrot, plasma)
So, it's not like a normie would think "Oh those AIs, I can ask it to build me complete software and it will do it instantly". There is something that is not been told when people say "yesterday this thing one-shooted". This is also why there is disagreement with programmers who used these tools extensively, half of them will say "It's magic, it does what I tell it and can one-shoot wholle big projects" and other programmers say "Bullshit, I have to babysit it for 1000 times".
Also there is a tendency in the social media to do exagerrated posts. Esp. with AI.
But I wonder, how long did it take? Does the agent work on things and self checks if everything is correct, like going in a feedback loop till it gets it right? How long does that take? Are there some unit tests or something?
Of course if it finished I see you ask it to change things and then it goes again.
Also,. if I understand, there are a lot of examples of previous effects which code might it be reusing. So, it's not like from a zero state it wrote algorithms for a lot of these effects without some previous examples? It's looking at the previous codes and grabs things and scripts them together, maybe change parameters?
The question is. Because this one-shoot can be misleading, people will think "If I go now on ChatGPT or even the more professional code LLMs, and I say "Please make a cool demo"" or even give some more thorough description as a user, like "make demo with this and that effect and that music and that everything",. from what I know it's not gonna give you that. You are for pain.
But maybe one has to build sophisticated system, with previous prompts and previous examples of other demos and effects, have it base it in that codebase to build a variation of it? Some of the effects in this seems to be variation from previous (lines, mandelbrot, plasma)
So, it's not like a normie would think "Oh those AIs, I can ask it to build me complete software and it will do it instantly". There is something that is not been told when people say "yesterday this thing one-shooted". This is also why there is disagreement with programmers who used these tools extensively, half of them will say "It's magic, it does what I tell it and can one-shoot wholle big projects" and other programmers say "Bullshit, I have to babysit it for 1000 times".
Also there is a tendency in the social media to do exagerrated posts. Esp. with AI.
Also, the earlier stages how broken were they? Just some things that needed to improved visually, timings of parts being longer, some small visual bugs? Or simply breaking and crashing? Some part had horrible visual errors? Would they stand running as demos with start to finish and it was just visual improvements?
What I mean generally, I don't think I can just right now go to Claude with nothing, blank slate and say "Start and get SDL2 and make empty project with framebuffer, timer, etc.. and now populate it with demo effects and script them and write the credits and everything" and oneshot the everything. It has to be something that is based on previous codebases. From what I understand, if in the same github repo I had examples of effect code that works, previous demos with scripting, complete things,. then maybe it can copy/steal parts of the code, like take the complete plasma effect and just change paremeters. Maybe this way it's possible to one-shoot variations of demos using the similar building blocks.
But normies will think "I can ask the AI to build complete software and I will get it same as images".
Only if you have orchestrated enough base building blocks in a repository and had it reuse what's already there I think.
Also, would we call this coding or simply agent orchestration?
What I mean generally, I don't think I can just right now go to Claude with nothing, blank slate and say "Start and get SDL2 and make empty project with framebuffer, timer, etc.. and now populate it with demo effects and script them and write the credits and everything" and oneshot the everything. It has to be something that is based on previous codebases. From what I understand, if in the same github repo I had examples of effect code that works, previous demos with scripting, complete things,. then maybe it can copy/steal parts of the code, like take the complete plasma effect and just change paremeters. Maybe this way it's possible to one-shoot variations of demos using the similar building blocks.
But normies will think "I can ask the AI to build complete software and I will get it same as images".
Only if you have orchestrated enough base building blocks in a repository and had it reuse what's already there I think.
Also, would we call this coding or simply agent orchestration?
Quote:
Also, would we call this coding or simply agent orchestration?
The latter. At my work it's more prompting and checking the
Code than coding. You just prompt and worry not to run of
The tokens ;). And as a good developer you will loose
With good developer fluent with ai. And yeah I hate it too
But what todo.
