That's all I've found as well, but, personally, I find that a bit unclear, for a couple of reasons. First, are they saying that the game itself can use generative AI, but it can't be used in the development of the game? So that would mean that if the game itself generates random levels using a generative AI approach, that's allowed, but, if I were to use that same code to pre-generate and manually modify the levels, that wouldn't be allowed because I'm now using generative AI as part of the development process? I.e., I can create a game that itself is a generative AI, but I can't use that AI I've built as part of the development of a downstream game?
And, second, what counts as generative AI? A lot of people wouldn't include procedural generative techniques in that definition, but, AFAIK, there's no consensus on whether traditional procedural approaches should be described as "generative AI".
And a third thing is, if I use an IDE that has generative AI, even for something as simple as code completion, does that run afoul of the rule? So, if I used Visual Studio with its default IntelliCode settings, that's not allowed because it has a generative AI-based autocomplete?
> AFAIK, there's no consensus on whether traditional procedural approaches should be described as "generative AI"
Sure there is. "Generative AI" is just a marketing label applied to LLMs - intended specifically to muddy these particular waters, I might add.
No one is legitimately confused about the difference between hand-built procedural generation techniques, and LLMs.
That's not quite true though, right? Because diffusion models are also generative AI and they're not LLMs. Heck, they probably got disqualified, not for the use of an LLM, but for the use of a diffusion model.
So I think Gen AI is an umbrella. The question is, do older techniques like GANs fall under Gen AI? It's technically a generative technique that can upscale images, so it's generating those extra pixels, but I don't know if it counts.
There's not that much difference between diffusion models and other auto-regressive models (https://www.youtube.com/watch?v=zc5NTeJbk-k). But I'm of the opinion that Generative AI is a terrible umbrella term. It should include basically all of digital art if we take it seriously. The flood fill / paint bucket tool can be considered AI, any program using a search algorithm can be phrased in AI terms of a sense-think-act loop. Nevertheless I do understand what people tend to mean by it when they're raging. Right now it might best be defined in terms of workflow: a human uses natural language to describe what they want, and moments later a plausible image appears trying to match. This clearly separates it from every other tool in the digital artist's program, even many which one could arguably call generative AI. It also separates it from stock-photo/texture searches done externally to some art program, as those are done in a query language rather than natural language.