New AI rules @revision
category: general [glöplog]
Sorry for another boring AI thread.
I really hoped for more clarification with the new rules but it still left me confused. especially with regards to code.
So generative AI (which to my understanding in coding context includes all the things like Claude, Copilot etc.) are not allowed. That is fine for me. Even though by default (in work context) I have it enabled and use the autocomplete. I can - and did for the recent productions - just turn it off for a demo production.
Now how does this relate to third party libraries. I assume third party libraries that are using AI are ok? For the fun of it I checked the SDL repo because this was used in the Revision invitation.
They were at least experimenting with AI reviews in SDL. Some example:
https://github.com/libsdl-org/SDL/pull/12730 here copilot was requested for a review.
and later the suggestions were applied: https://github.com/libsdl-org/SDL/compare/895ce97042848a1c371dc3518ec573503e074500..91b33fd1f1406b8007ba36370a75c5c5376e3f40
Not directly but the suggestions from the review were definitely being used.
there are a few more references which imply usage of AI here and there.
I mean SDL is not an AI generated library and probably 99.9999% of the code is "gen-AI-free" but it shows how omnipresent the topic is in code is by now. I know I might be nitpicking but given the potential community backslashes that we had I think the rules should be crystal clear on what is allowed and what not.
So I assume that the revision rules would not ban their own invitation. I also don't think that unreal and unity are ruled out (nobody can tell me that they are not using some coding assistants somewhere during development).
so we are likely talking about my own code that I wrote for the demo and not about third party libraries and tools. if I were to use blueberrys no-cpu framework - which to my understanding from previous discussions was developed while copilot autopilot was on - to make a production that would be fine. however is blueberry allowed to use his own framework now? :D
another question. when googling nowadays you will get an AI summary and I guess even the best coders will have to google something somewhen. copy pasting the result: yes or no? I am not even going down the road that some answers on stack overflow might just be AI content.
for "pong-rage" I was using google mediapipe during preprocessing. no LLM but according to the wikipedia page about generative AI "These models learn the underlying patterns and structures of their training data, and use them to generate new data". this seems to rule out google mediapipe?
also during development of "rise of kotzilla" we evaluated cascadeur (https://cascadeur.com/). we didn't use it in the end but this is an AI first tool. it is not about "generate me a nice dance animation" but the AI models in there help for in-betweening etc. here is a nice interview from the CEO: https://80.lv/articles/cascadeur-talks-inbetweening-ethical-ai-training-the-future-of-the-ai-vs-anti-ai-debate. yes or no? again. the models generate information from their training data which again matches the definition from wikipedia.
I am also not a musician but I guess there are tons of newer samples that might also be AI generated? Even if not disclosed and the samples are free to use and so are fine to be used @revision. I mean this is likely ok. how would you be able to check that.
AI is so omnipresent wether you like it or not that I think some examples on the rules page would be really helpful to clarify things :(
I really hoped for more clarification with the new rules but it still left me confused. especially with regards to code.
So generative AI (which to my understanding in coding context includes all the things like Claude, Copilot etc.) are not allowed. That is fine for me. Even though by default (in work context) I have it enabled and use the autocomplete. I can - and did for the recent productions - just turn it off for a demo production.
Now how does this relate to third party libraries. I assume third party libraries that are using AI are ok? For the fun of it I checked the SDL repo because this was used in the Revision invitation.
They were at least experimenting with AI reviews in SDL. Some example:
https://github.com/libsdl-org/SDL/pull/12730 here copilot was requested for a review.
and later the suggestions were applied: https://github.com/libsdl-org/SDL/compare/895ce97042848a1c371dc3518ec573503e074500..91b33fd1f1406b8007ba36370a75c5c5376e3f40
Not directly but the suggestions from the review were definitely being used.
there are a few more references which imply usage of AI here and there.
I mean SDL is not an AI generated library and probably 99.9999% of the code is "gen-AI-free" but it shows how omnipresent the topic is in code is by now. I know I might be nitpicking but given the potential community backslashes that we had I think the rules should be crystal clear on what is allowed and what not.
So I assume that the revision rules would not ban their own invitation. I also don't think that unreal and unity are ruled out (nobody can tell me that they are not using some coding assistants somewhere during development).
so we are likely talking about my own code that I wrote for the demo and not about third party libraries and tools. if I were to use blueberrys no-cpu framework - which to my understanding from previous discussions was developed while copilot autopilot was on - to make a production that would be fine. however is blueberry allowed to use his own framework now? :D
another question. when googling nowadays you will get an AI summary and I guess even the best coders will have to google something somewhen. copy pasting the result: yes or no? I am not even going down the road that some answers on stack overflow might just be AI content.
for "pong-rage" I was using google mediapipe during preprocessing. no LLM but according to the wikipedia page about generative AI "These models learn the underlying patterns and structures of their training data, and use them to generate new data". this seems to rule out google mediapipe?
also during development of "rise of kotzilla" we evaluated cascadeur (https://cascadeur.com/). we didn't use it in the end but this is an AI first tool. it is not about "generate me a nice dance animation" but the AI models in there help for in-betweening etc. here is a nice interview from the CEO: https://80.lv/articles/cascadeur-talks-inbetweening-ethical-ai-training-the-future-of-the-ai-vs-anti-ai-debate. yes or no? again. the models generate information from their training data which again matches the definition from wikipedia.
I am also not a musician but I guess there are tons of newer samples that might also be AI generated? Even if not disclosed and the samples are free to use and so are fine to be used @revision. I mean this is likely ok. how would you be able to check that.
AI is so omnipresent wether you like it or not that I think some examples on the rules page would be really helpful to clarify things :(
It should be clarified, but the gist of it is "don't use AI for art and don't vibecode".
As in, *you* shouldn't make your art with AI, and *your* code shouldn't be AI.
Copilot is so dumb, that it doesn't count IMO.
It's not a political stance against AI - it's about keeping the humanness in art and you are the human.
As in, *you* shouldn't make your art with AI, and *your* code shouldn't be AI.
Copilot is so dumb, that it doesn't count IMO.
It's not a political stance against AI - it's about keeping the humanness in art and you are the human.
While I agree that the rules are problematic and ultimately unenforceable when it comes to code (even by contestants themselves in good faith), I do think they are rather clear as such compo rule clauses go. They avoid mushy and ill-defined formulations like "usage of AI" and instead refer specifically to AI-generated content. I would like to commend the Revision organizers for this clarity.
This formulation also implies that using AI code generation for tooling is within the rules. I think this is a very reasonable and productive place to draw the line.
Actually, from what I have heard from people working at Epic Games, they completely ban AI code generation and assistance for all code that ends up in the public Unreal Engine codebase, for precisely this reason.
The No-CPU framework would be permitted, since the code exists purely on the tooling side and none of it ends up in the demo.
This formulation also implies that using AI code generation for tooling is within the rules. I think this is a very reasonable and productive place to draw the line.
Quote:
So I assume that the revision rules would not ban their own invitation. I also don't think that unreal and unity are ruled out (nobody can tell me that they are not using some coding assistants somewhere during development).
Actually, from what I have heard from people working at Epic Games, they completely ban AI code generation and assistance for all code that ends up in the public Unreal Engine codebase, for precisely this reason.
Quote:
so we are likely talking about my own code that I wrote for the demo and not about third party libraries and tools. if I were to use blueberrys no-cpu framework - which to my understanding from previous discussions was developed while copilot autopilot was on - to make a production that would be fine. however is blueberry allowed to use his own framework now? :D
The No-CPU framework would be permitted, since the code exists purely on the tooling side and none of it ends up in the demo.
I'm wondering if I can use browser's speech API in my possible entries. The rules for browser demos do say that "All technologies natively supported by the browser are permitted", but on the other hand speech API delegates the audio generation to the OS's text to speech, which in many cases (e.g. macOS and I think Windows too) uses neural network based synthesis provided by the operating system itself.
Quote:
I'm wondering if I can use browser's speech API in my possible entries. The rules for browser demos do say that "All technologies natively supported by the browser are permitted", but on the other hand speech API delegates the audio generation to the OS's text to speech, which in many cases (e.g. macOS and I think Windows too) uses neural network based synthesis provided by the operating system itself.
As I read the rules, this would be allowed, since your demo doesn't contain the AI-generated content; it merely accesses it via a resource that it is explicitly permitted to access.
One possible further clarification to the rules could be what exactly is meant by AI-generated content. Does it just refer to content produced by a model that has been trained on material that you do not have a license for (true for basically all the big models you can use/download), or does it also cover cases where you own the entire training material? I would assume the former. If my memory serves me right, there have been several productions of the latter kind in the past.
Imho drawing the line is quite simple - if you have to ask, the answer is NO
I just wanted to raise this question since the OS text to speech has been exploited in demos probably as long it has been possible. Especially in size coded demos it's used to impress the audience, that you can have also speech in such a tiny executable. I've also seen at least one recent web demo, who's audio was the source code of the demo itself spoken aloud using TTS.
what groepaz said
Actually not what Groepaz said but pretty much what Wrighter said.
So no, you don't need to vet every single 3rd party library or API use for possible AI contributions, as long as you're within the rules otherwise. And as Blueberry said, vibe coded/assisted tooling such as the no-CPU framework is allowed as long as it's not part of the prod itself. Using AI generated tools isn't that much different than using a 3rd party engine (arguably even less bad), and we're coping with the latter just fine-ish as a scene, depending on who you ask.
And if you're training your own ML model on your own data and then make a demo with/about it, that's fine too.
But I'll be frank: I don't think it's possible to formulate a reasonably sized set of rules on the topic that won't raise any questions or incentivize people to look for loopholes, and in the latter case, please understand that we as orgas have better things to do than deal with any instance beforehand.
In the end, ask yourself: "did I make the thing, or did I have the thing made for me?" because that's the gist of it, really. If you act in good faith, you'll probably not get disqualified. And if you're unsure, our hotline email is always open.
So no, you don't need to vet every single 3rd party library or API use for possible AI contributions, as long as you're within the rules otherwise. And as Blueberry said, vibe coded/assisted tooling such as the no-CPU framework is allowed as long as it's not part of the prod itself. Using AI generated tools isn't that much different than using a 3rd party engine (arguably even less bad), and we're coping with the latter just fine-ish as a scene, depending on who you ask.
And if you're training your own ML model on your own data and then make a demo with/about it, that's fine too.
But I'll be frank: I don't think it's possible to formulate a reasonably sized set of rules on the topic that won't raise any questions or incentivize people to look for loopholes, and in the latter case, please understand that we as orgas have better things to do than deal with any instance beforehand.
In the end, ask yourself: "did I make the thing, or did I have the thing made for me?" because that's the gist of it, really. If you act in good faith, you'll probably not get disqualified. And if you're unsure, our hotline email is always open.
Quote:
"did I make the thing, or did I have the thing made for me?"
Extremely well said. This should probably be added to the partysystem's entry submission page :)
"Did I make the thing, or did I have the thing made for me?" actually misses the point. You can make a stunning demo in UE without knowing the technical aspects behind it, so the entry level is already pretty low. And if you don't want to use 3rd parties, just use a premade intro framework, slap on a refurbished shader, and add your artistic spin.
So, why does using AI hit differently? I can only speculate. Maybe because it's inherently non-demoscenic and risks attracting people who care only about final results, not the craft that the scene has built up over decades.
And yet, that doesn't stop people from experimenting. Interestingly, those experimenting are not complete noobs. So why not give them the benefit of the doubt?
So, why does using AI hit differently? I can only speculate. Maybe because it's inherently non-demoscenic and risks attracting people who care only about final results, not the craft that the scene has built up over decades.
And yet, that doesn't stop people from experimenting. Interestingly, those experimenting are not complete noobs. So why not give them the benefit of the doubt?
