Rick Rubin on AI (& now Graeme Revell, too)

Anything about MUSIC but doesn't fit into the forums above.
Post Reply New Topic
RELATED
PRODUCTS

Post

An author is paid each time a book is taken out of a library. In the UK it is not very much and I believe there is also an annual cap for each title.

Post

That is absolutely correct but you don't pay for that, even if you use the knowledge and experience gained to make a million dollars, nor do you pay for pretty much anything you can access on the internet. Companies shouldn't have to pay for their AIs to access that information either. It's basically open source. Let's use KVR as an example - beyond the forums there is a wealth of data available on this site, open to anyone who wants it, all for free. Why shouldn't AI be able to train on this data for free, just like the rest of us?

If access requires a free account, then by all means make them create an account for their AI. If there is a subscription fee, then the AI company should pay for a subscription if they want access to the content but beyond that, any other monetary expectation is pure greed.
v4p0r wrote: Sun Feb 01, 2026 12:23 pmThey should be licensed because they are created without understanding.
You were also created without understanding but the state provided you with a free education, at great cost to taxpayers. Training AI doesn't impose any costs on those providing the training data so why should it have to be paid for?
If I gave instructions to an artist to draw three frames of an object acting in a certain way, and then I get back only two frames, the first one and the third one, and I never can instruct him to draw the second one, I simply must conclude that he cannot draw. And when he cannot draw, he probably gave me copies. And then I would need a license for those two copies.
This makes absolutely no f**king sense in the context of the question, even taking into account the fact that English may not be your first language. It's nonsensical.
NOVAkILL : Legion GO, AMD Z1x, 16GB RAM, Win11 | Audient EVO 8 | Lumi Keys | Studio Pro 8
Korg Odyssey, bx-oberhausen, Proxima, PolyMax, GR8, JP6K, Union, Atomika,
Invader 2, Flow Motion, Olga, TRK 01, Thorn, Spire, VG Iron

Post

Ok, you wanted the library question to be answered. The general public has paid already for a lot of existing content, and AI companies provide great value. I agree as stated earlier, I get value from it - be it just in a different way then before AI existed.
I don't think the fair-use question is easy to answer, because the laws have been made for humans, and humans memorize content differently. If there come new laws that also clarify in what cases I can use the AI agents output, then both the AI companies and the customers are helped by this. Whether it should still involve licensing isn't really my concern, I just thought it might be fair to have an additional regulation for that, but I don't want to make life hard for Open AI after receiving value in the form of ideas and inspiration, if that's the concern.

Post

If you're worried about AI, this should scare the living shit out of you -

NOVAkILL : Legion GO, AMD Z1x, 16GB RAM, Win11 | Audient EVO 8 | Lumi Keys | Studio Pro 8
Korg Odyssey, bx-oberhausen, Proxima, PolyMax, GR8, JP6K, Union, Atomika,
Invader 2, Flow Motion, Olga, TRK 01, Thorn, Spire, VG Iron

Post

Moltbook is like telling fraudsters "shut up and take my money". You might just as well publish your bank details and passwords on Reddit. It's the mother of cross-site scripting attacks.

Post

AI is bad. Just wait until you have to try to release something new and try to promote it. Just about every platform for new artist releases is flooded by AI slop. The only way to complete is to make a 1,000 tracks of AI slop and compete with other AI slop.

Post

I think it's great. The more AI slop, more ends up in training data which causes model collapse. The slop also makes real artists stick out much more.

Post

Zeisner wrote: Mon Feb 02, 2026 3:19 pm I think it's great. The more AI slop, more ends up in training data which causes model collapse. The slop also makes real artists stick out much more.
AI gets trained on AI which gets trained on AI...

Wonder what it would sound like after, say, 100 iterations?

Post

Bunny_boy wrote: Mon Feb 02, 2026 4:21 pm
Zeisner wrote: Mon Feb 02, 2026 3:19 pm I think it's great. The more AI slop, more ends up in training data which causes model collapse. The slop also makes real artists stick out much more.
AI gets trained on AI which gets trained on AI...

Wonder what it would sound like after, say, 100 iterations?
It would sound like what all music sounds like for a run-of-the-mill, billionaire tech-bro: collection of sounds that fill the voids of lives, driving people to hold on to their alienation more and more, and by that function makes the tech-bro amass even more capital, which in turn will bring even more capital, and so on.

Post

csscp wrote: Mon Feb 02, 2026 4:29 pm
Bunny_boy wrote: Mon Feb 02, 2026 4:21 pm
Zeisner wrote: Mon Feb 02, 2026 3:19 pm I think it's great. The more AI slop, more ends up in training data which causes model collapse. The slop also makes real artists stick out much more.
AI gets trained on AI which gets trained on AI...

Wonder what it would sound like after, say, 100 iterations?
It would sound like what all music sounds like for a run-of-the-mill, billionaire tech-bro: collection of sounds that fill the voids of lives, driving people to hold on to their alienation more and more, and by that function makes the tech-bro amass even more capital, which in turn will bring even more capital, and so on.
Oh it would be horrible, totally. But something that is inherintly horrible, and quite removed from human interaction (apart from the remnants of the original songs and the coders' biases). Might be interesting, might be soul destroying. Might go with the second of those myself!

Post

Bunny_boy wrote: Mon Feb 02, 2026 4:21 pm AI gets trained on AI which gets trained on AI...

Wonder what it would sound like after, say, 100 iterations?
Total randomness, noise. OpenAI is already having trouble with model collapse, they had to do two rollbacks last year because ChatGPT went bonkers. AI is already on life support. It needs plenty of new data as fast as possible to slow down model collapse - which is one if the reasons why AI gets shoved into operating systems (especially Windows). To record and send everything users do. Think about all the data that would never be uploaded, all the demos and takes in your DAW.

Post

Bunny_boy wrote: Mon Feb 02, 2026 4:32 pm Oh it would be horrible, totally.
Noise (Both white and pink) seems to be rather soothing for humans so that's actually not that bad. A little expensive and resource-hungry though when I compare it to mda Test Tone...

Post

csscp wrote: Mon Feb 02, 2026 4:29 pm
Bunny_boy wrote: Mon Feb 02, 2026 4:21 pm
Zeisner wrote: Mon Feb 02, 2026 3:19 pm I think it's great. The more AI slop, more ends up in training data which causes model collapse. The slop also makes real artists stick out much more.
AI gets trained on AI which gets trained on AI...

Wonder what it would sound like after, say, 100 iterations?
It would sound like what all music sounds like for a run-of-the-mill, billionaire tech-bro: collection of sounds that fill the voids of lives, driving people to hold on to their alienation more and more, and by that function makes the tech-bro amass even more capital, which in turn will bring even more capital, and so on.
like general midi dire straits.
:ud:

Post

Gamma-UT wrote: Mon Feb 02, 2026 12:44 pmMoltbook is like telling fraudsters "shut up and take my money". You might just as well publish your bank details and passwords on Reddit. It's the mother of cross-site scripting attacks.
Except it's not for people or any kind of entity with a bank account. What are you going to steal from an AI agent that anyone would care about? The whole concept is fascinating to watch.
nitrateaudio wrote: Mon Feb 02, 2026 2:01 pmAI is bad. Just wait until you have to try to release something new and try to promote it. Just about every platform for new artist releases is flooded by AI slop. The only way to complete is to make a 1,000 tracks of AI slop and compete with other AI slop.
How is that any different to two or three years ago? The difference that AI has made to barriers of entry is insignificant compared to the institutional barriers that have always been in place.

The most successful thing I ever released was my first ever record, a 7" single that was released in 1987. Back then I could send a video tape to the ABC (Australian equivalent of the BBC) and get it played on one of their weekly music shows. I could get distribution from any of a dozen or more local, independent distributors that put it into record shops where people could easily find it and buy it. I could drop copies to ABC radio and get airplay on a radio station people actually listened to. I was even able to get a promotional spot on the local TV news. 35 years later, random people from the other side of the world started uploading it to YouTube and 10 years after that, I was approached by a Belgian DJ who wanted to kick of his new label by re-releasing it as a 12", which he did. None of that happened because the single was any good, it wasn't, it happened because it was so much easier to make shit happen back then.

Everything is so diffuse now that it's impossible. AI may have made that incrementally more difficult but 99% of the damage was done years ago.
Zeisner wrote: Mon Feb 02, 2026 3:19 pmI think it's great. The more AI slop, more ends up in training data which causes model collapse. The slop also makes real artists stick out much more.
What sort of moron would allow their AI to train on "slop"? "Garbage in, garbage out" has been an adage of the computer age for 50 years or more, the people making the AIs understand this.
Bunny_boy wrote: Mon Feb 02, 2026 4:21 pmAI gets trained on AI which gets trained on AI... Wonder what it would sound like after, say, 100 iterations?
Keep watching Moltbook and you'll probably find out in a week or two.
csscp wrote: Mon Feb 02, 2026 4:29 pmIt would sound like what all music sounds like for a run-of-the-mill, billionaire tech-bro: collection of sounds that fill the voids of lives, driving people to hold on to their alienation more and more, and by that function makes the tech-bro amass even more capital, which in turn will bring even more capital, and so on.
Maybe if the tech bro is doing the prompting. How successful do you think an AI will be that doesn't do what the prompter, the person paying for the service, tells it to do?
Bunny_boy wrote: Mon Feb 02, 2026 4:32 pmOh it would be horrible, totally. But something that is inherintly horrible, and quite removed from human interaction (apart from the remnants of the original songs and the coders' biases).
How f**king stupid are you people? AI does what you tell it to, not what someone else wants it to. How is that not completely f**king obvious?
Zeisner wrote: Mon Feb 02, 2026 5:36 pmAI is already on life support.
Keep telling yourself that, Chicken Little. The thing with LLMs is that they aren't the end-point of AI. We're already seeing that Hierarchical Reasoning Models (HRMs) are outperforming LLMs. LLMs are just the current stage, AI is only going to get better and better until it leaves us all far behind. Get used to the fact that your obsolescence is guaranteed and will probably be happening sooner rather than later, if it hasn't happened already.
NOVAkILL : Legion GO, AMD Z1x, 16GB RAM, Win11 | Audient EVO 8 | Lumi Keys | Studio Pro 8
Korg Odyssey, bx-oberhausen, Proxima, PolyMax, GR8, JP6K, Union, Atomika,
Invader 2, Flow Motion, Olga, TRK 01, Thorn, Spire, VG Iron

Post

BONES wrote: Tue Feb 03, 2026 12:29 am Everything is so diffuse now that it's impossible. AI may have made that incrementally more difficult but 99% of the damage was done years ago.
If AI succeeds, it might make it it 99,999% times more difficult to get anything heard out there. AI artists making AI music for AI bots.

Post Reply

Return to “Everything Else (Music related)”