Forums

Well, you know I like principles. I'd just personally have a hard time writing that specific principle up, in a universal manner, without it essentially being distilled into Luddism. History is full of examples of technology making some jobs obsolete, or much easier, and I kinda think we're on the first of many waves of AI basically doing that; while I hope that we can get governance and social safety nets in place ahead of time, my observation is that most regulations are reactive, not proactive, in this regard. Perhaps you can rephrase what you think the generalized principle actually is, in a way that wouldn't essentially prevent any new technology from ever disrupting any job market? I believe it's difficult? Sure, I think requesting a brief statement on the use of AI, if it was employed - even if that statement is written by AI, so long as it is accurate - is a decent way of trying to capture the influence of the tech on any given track, in an honor-system mode. Ala Valve. I'm not convinced that usage in a large training dataset, where aesthetics/characteristics are gleaned and end up in a kind of weird soup that almost reminds me of human brains at times, is directly equivalent to usage as a "mere asset". People use OC ReMixes in far LESS transformative fashion, where I suppose you could call it a "mere asset",  when they employ them in non-profit works, streams, let's play videos, etc., don't they? The goal of promoting VGM has some siblings - one related goal is music education, and I think we've always hoped that folks would learn from arrangements, study them, be inspired by them, etc. I guess the key word there is "folks" - it's definitely not the same with AI, because of the scale & speed. I don't want to sound defeatist, here, but I also want to repeat that there is enough music that is completely in the public domain that a pretty capable model could be built from JUST that corpus, so it becomes unclear what is being achieved, precisely... I think this is a great, essential conversation to be having, and this announcement does actually seem like the relevant place for it...

Added to that: be careful that the arrangement still has to be considered a cover, and if you’re very liberal with your arrangement, that may become a problem. Not that I’ve seen that myself, but I already vet the ones that I think are too liberal and simply don’t try to post them. Another big one: original lyrics are a no-go. If the source track has lyrics, you can use those, but anything with your own lyrics won’t fly. To be safe, also for stuff like choirs and so on, stick with the regular oohs and aahs (even though you might possibly get away with the single phrase here and there).

THIS IS SO FUCKING COOL! Really creative. Bold choice to leave everything so sparse, but it works spectacularly, and really gives the tablas room to breathe throughout the mix. Definitely going on the favorites list.

I think, in principle, it's not so much the training data that's the ethical problem but how the network can be used to replace the people who made the training data. In our case, musicians. Remixers. Composers. Much like how using images on the internet to create a tool to help blind people see is imo fine, but a tool to make photographers and illustrators largely obsolete isn't. I don't think it's in line with ocr's goal of promoting vgm as an art form to let people use remixes as mere assets for other works, including future vgm, so I think it'd be good to communicate this, even if it's unenforceable. A principle. I like how valve wants disclosure of how AI is used in the development and functionality of games on steam.

Also, FWIW, you are right in some regard on the length alone that it was too much of a slow-roll and overambitious to a fault. ...I don't know. It was a darling I wasn't willing to kill, considering some background circumstances / context behind this remix from the album itself that hits pretty close to home. ...It was a vision indeed, but not without its flaws. I'll just do better next time.