AI tooling in the Demoscene
category: code [glöplog]
Based on the many discussions I had at Revision this year it seems there's a topic many have on their minds, and that is to build the tooling for productions with the help of the new agentic coding models.
Gopher had a talk on the topic and I myself had an experiment for an SDF editor that I then used to create my 4k exegraphics entry.
The consensus so far seems to be that using these tools to build the boring and tedious parts of tooling (boilerplate, third party integration, ui, you know the things you hate) is a huge boon. At the end of his talk Gopher posed a couple of poignant questions along with "Am I no longer a demoscener now?".
In many situations on the scene where new technology could be used to overtake people who have been honing their craft for decades towards very specific skills the result was a very loud uproar (see the pixel vs scan debates or even sw vs hw rendering as examples). Based on this, with ~24 years of tool development behind me I should be one of the biggest opponents of utilizing AI for someone to make in half a year that took me decades to master. The way I see it though is that if this for example lowers the entry bar for building 64k productions I'm all for it.
The thing I learned from "creating" the SDF editor is that these things won't help you much (yet) if you don't know how you'd build something in the first place. And the code was generated is of such low quality that I'd never release it and have my name associated with it - but on the other hand it did bring out all the initial growing pains in tooling that I'm used to seeing and fix in the next iteration. An iteration that I'd build manually based on all that I have learned (with what I consider to be a throwaway prototype) and that could be used as a long term solution.
Throughout the party basically any coder I talked with brought up this same topic. It's clearly of interest and we as a scene need to decide where we stand on it.
And before anyone derails the discussion: this is about tooling and not content (including released effect code) creation/generation. That's a very different topic with what I suspect a very different outcome.
Discuss.
Gopher had a talk on the topic and I myself had an experiment for an SDF editor that I then used to create my 4k exegraphics entry.
The consensus so far seems to be that using these tools to build the boring and tedious parts of tooling (boilerplate, third party integration, ui, you know the things you hate) is a huge boon. At the end of his talk Gopher posed a couple of poignant questions along with "Am I no longer a demoscener now?".
In many situations on the scene where new technology could be used to overtake people who have been honing their craft for decades towards very specific skills the result was a very loud uproar (see the pixel vs scan debates or even sw vs hw rendering as examples). Based on this, with ~24 years of tool development behind me I should be one of the biggest opponents of utilizing AI for someone to make in half a year that took me decades to master. The way I see it though is that if this for example lowers the entry bar for building 64k productions I'm all for it.
The thing I learned from "creating" the SDF editor is that these things won't help you much (yet) if you don't know how you'd build something in the first place. And the code was generated is of such low quality that I'd never release it and have my name associated with it - but on the other hand it did bring out all the initial growing pains in tooling that I'm used to seeing and fix in the next iteration. An iteration that I'd build manually based on all that I have learned (with what I consider to be a throwaway prototype) and that could be used as a long term solution.
Throughout the party basically any coder I talked with brought up this same topic. It's clearly of interest and we as a scene need to decide where we stand on it.
And before anyone derails the discussion: this is about tooling and not content (including released effect code) creation/generation. That's a very different topic with what I suspect a very different outcome.
Discuss.
i simply just switched to c64. That more-or-less solves the issue at least for a few years.
I’m fully on board with using AI to generate boilerplate, integration, tooling code. Writing the same editor window for the nth time isn’t particularly engaging.
I absolutely dislike AI for generating art, text, music, ie. "content". The art I want to see is self-expression from people.
An editor window, a build system or a bunch of boilerplate are not self-expression. No one cares about them. They're a mandatory evil. I've been trying out some AI tools at work and they've been really helpful. If it helps with more demos coming out, I'm all for it.
An editor window, a build system or a bunch of boilerplate are not self-expression. No one cares about them. They're a mandatory evil. I've been trying out some AI tools at work and they've been really helpful. If it helps with more demos coming out, I'm all for it.
My take is that AI currently is much more useful for analysis than it is for generation. I'm biased since I genuinely enjoy writing all sorts of code by hand, but I think writing your own boilerplate-laden "plumbing" code is probably actually good for your long term brain health, just like doing crosswords or sudokus. But I guess it evens out in the long run as long as you keep engaging your brain, and I'm not against anyone using tooling for automating things they find tedious; we're all different and I wouldn't want others telling me which tools I must use or refrain from using.
I'm neutral towards it as a technology, in conclusion. The companies who are currently busy doing the shoving of LLMs down our throats are a different matter, and I won't stand for actively or passively supporting their ilk.
I'm neutral towards it as a technology, in conclusion. The companies who are currently busy doing the shoving of LLMs down our throats are a different matter, and I won't stand for actively or passively supporting their ilk.
Quote:
utilizing AI for someone to make in half a year that took me decades to master
Is that a likely outcome though?
Quote:
If it helps with more demos coming out, I'm all for it.
Well i guess the erosion of democracy, the rise of fascism and the destruction of the environment are a small price to pay if we get a few more demos out of it.
Quote:
Quote:utilizing AI for someone to make in half a year that took me decades to master
Is that a likely outcome though?
I already know of someone thinking of using agentic AI to work with one of our older released codebases. New tooling from scratch based on released code isn't a stretch imo.
Let's agree on definitions here first. Officially, all of those models are considered genAI (fact-check yourself): audio, video and text models (LLMs). It's just the different medium they generate. But the principle is similar: collect as much as data from the public domain ignoring licensing at large, train a model that learns universal transformation formulas and not the content itself (hence a legal loophole).
I don't buy the argument artists are more important than coders.
As I shared in one-liner it's very possible to make an entry without any human input (256b compo idea, nothing surprising ofc) and you can expect non-human entries to only improve over time.
So the question to me boils down what is the ratio of AI vs human work that is acceptable across all media (code, audio, visuals). Tooling is very much part of it as well (lies at intersection between code and artistic content creation), but just harder to police (no visible artefacts in a demo itself).
I also cannot resist to say, it's ironic how you guys change your mind if things suits *you*.
I don't buy the argument artists are more important than coders.
As I shared in one-liner it's very possible to make an entry without any human input (256b compo idea, nothing surprising ofc) and you can expect non-human entries to only improve over time.
So the question to me boils down what is the ratio of AI vs human work that is acceptable across all media (code, audio, visuals). Tooling is very much part of it as well (lies at intersection between code and artistic content creation), but just harder to police (no visible artefacts in a demo itself).
I also cannot resist to say, it's ironic how you guys change your mind if things suits *you*.
Quote:
Well i guess the erosion of democracy, the rise of fascism and the destruction of the environment are a small price to pay if we get a few more demos out of it.
I'm all for burning down the companiens and banning the fucking things for all eternity and I wish they didn't exist, but it is there on my VS and it's helping me write boring code. Me using it for debugging a random crash won't be the downfall of democracy.
Hypocritical? I guess.
I think the use for tooling is fine, but I think we also have to talk about actual potentially generated content. To scope this a bit: Let's maybe restrict ourselves to size coding - 256b, 4k, 4k exe gfx, 8k, 64k. For me generated images/music somehow feel like a very different thing than code and I definitely don't want to be involved in the debate about "is it fine to completely generate my demo soundtrack with AI". One problem at a time. I think if we exclude generated images/music just for the sake of the debate, we should be able to find at least some reasonable consensus.
Compo organizers literally have absolutely zero chance to check whether you generated the code for your intro using any of these tools. Why should we especially disqualify people that are actually honest about their ML-tool use? I think that's ridiculous.
The reality of software development is changing and to think that that's going away anytime soon is simply naive. None of the extreme positions help here. You don't have to like the technology to understand that it's not simply going away. Oh and it affects the demoscene.
Besides the "killing the planet issues", which are obvious but not really helping in the debate (make a demo about it, thx). How do we deal with potential "pay to win"? Is that OK? Do we care? The good stuff costs actual money and it's not a "buy once cry once". That's not really super accessible for everyone. Not a fan of that.
The rules have to be a thing that is shaped by the people who do the actual productions. So what should it be? If the consensus is complete escalation and allowing ML stuff in all PC size coding competitions. Why not?
Compo organizers literally have absolutely zero chance to check whether you generated the code for your intro using any of these tools. Why should we especially disqualify people that are actually honest about their ML-tool use? I think that's ridiculous.
The reality of software development is changing and to think that that's going away anytime soon is simply naive. None of the extreme positions help here. You don't have to like the technology to understand that it's not simply going away. Oh and it affects the demoscene.
Besides the "killing the planet issues", which are obvious but not really helping in the debate (make a demo about it, thx). How do we deal with potential "pay to win"? Is that OK? Do we care? The good stuff costs actual money and it's not a "buy once cry once". That's not really super accessible for everyone. Not a fan of that.
The rules have to be a thing that is shaped by the people who do the actual productions. So what should it be? If the consensus is complete escalation and allowing ML stuff in all PC size coding competitions. Why not?
Quote:
…these things won't help you much (yet) if you don't know how you'd build something in the first place…
I’m slightly offtopic but bear with me please. I agree with that quote. Pay attention to “(yet)”. Do you see the writings on the wall? Do you feel it? Do you really want to squeeze as much demos as you can from yourselves and let it all die with you? Let it all burn? You are willing to sacrifice the whole future of this subculture for a few more demos before you die? Are you nuts? And I’m not talking about the environment now (although we should), I’m talking about the possibility of the next generation of coders not being able to get things done without AI writing the boilerplate and/or tooling. Now, in the mainstream, that’s just the natural progression, this abstraction upon abstraction, after all the first coders ran around with tubes and pushed switches. English is a language, just like Java. But in this subculture? In this subculture?! You want to totally abstract away the coding part in a subculture that’s all about coding? Or atleast “the boring part” as you call it. Well, look, demos are audiovisual works. Bearing that in mind, one may say ALL coding is the boring part of audiovisual creation. But then we’re off to motion gfx/animation/sfx. And demoscene loses all meaning. Why should it even exist then?
