Bowser: “We still believe that what makes our games special is our developers, their artistic capabilities, their insight into how people play. So, there’s always, always going to be a human touch, and a human engagement in how we develop and build our games.”
How did you come up with that? There’s no discussion on whether Nintendo sources training data ethically / legally. They could train it on their own property since they own tons too. Given Japanese sensibilities I’d be surprised if Nintendo behaved like Western companies in that regard and that’s part of Nintendo quirkiness that I will defend like a fanboy.
The algorithms don’t just appear by themselves, and they have to be refined upon something. Are you telling me you honestly believe Nintendo is going to roll their own AI using ethically sourced art and text? Because I don’t.
Nintendo creates art and text for decades now mate. They don’t need to steal, they can cross-license that stuff with other IP holders for even more content to train on. You must be thinking of parasitic GenAI companies that do not create anything but bundled Nintendo with them because you don’t like them.
You missed the point. I’m not talking about just giving a model their library of artwork, which they are within their rights to do.
That model that can parse their artwork had to be developed and refined upon other work. Unless Nintendo is building their own upon their artwork—which I seriously doubt, because that would be a costly undertaking—they’ll be using existing models licensed from somewhere, all of which were developed by stealing from legitimate artists. If they’re using it for idea generation (like an LLM), that text generator was built upon stolen writing.
I’m merely pointing out that it’s ironic for a company that takes such a harsh stance with its fans for “stealing” its IP to be okay with AI.
What you’re describing is fine tuning, not model creation.
You can train diffusion models from scratch, even on home hardware, using open source software. It is well with the capability of Nintendo to do this with their own artwork.
Adobe did, they created their models from artwork licensed from artists specifically for training their models.
There’s no reason to think that Nintendo would use public diffusion models when they can train their own and have a model that more accurately reflects their style.
You assume too many things you don’t know, that’s no way to come to reasonable conclusions unless you mean to come to predefined notions.
I’m okay with a little irrationality this morning. Nintendo can eat shit.
They have plenty of artwork to make their own Bowser model, or Mario model. Llms and image generators are AI, AI is not only LLMs and image generators. Anyone can create a new model of content they own, and they have plenty.
Cool, and you think Nintendo is spending the resources to develop their own model from scratch? Because if not, the existing models were built and refined upon other people’s art (written, drawn, etc.)
I don’t think they are, because that would be costly; it’s much easier to just enter a licensing agreement.
Azure, AWS, and GCP all literally have tools to drag and drop and train a model.