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

[csscp]

It's dumb

Do the following equation in seconds: 2,673 + 4,654 + 5,212 x 3 + 8,789.
Ok times up. You failed. You're dumb.
But guess what? A calculator can do it.

If your idea of intelligence is being able to do 4+ digit arithmetic rapidly, then, sure. But a world where your definition of intelligence is common is already destined to be a mindlessly rigid, robotic, barbaric, self-destructive place, which is very much like our world today.

[VOODOO U]

If your idea of intelligence is being able to do 4+ digit arithmetic rapidly, then, sure. But a world where your definition of intelligence is common is already destined to be a mindlessly rigid, robotic, barbaric, self-destructive place, which is very much like our world today.

No. The "world" is not a mindlessly rigid, robotic, self -destructive place. Human beings are mindlessly rigid, robotic and self-destructive.
When we're gone, the world will carry on.

[BONES]

But were the brains you're using trained on anything but perfection?

Yes, anything but perfection. There is a huge amount of randomness in the process. As I've explained half-a-dozen times already, if you feed it the same prompt 10 times, you'll get 20 very different results. Anything you don't specify, it will throw random stuff into.

Think of it a bit like a random note generator (which I use a lot when I'm evaluating new instruments). The one in Orion had an algorithm that helped ensure you always got something musical from it. If you pressed CTRL+R enough times, you'd end up with a usable riff. There's a song on our first album where the bassline in the verse is exactly that. AI is kind of the same, if you keep asking it to generate something base don the same prompt, it will eventually give you something you're happy with. If it was perfect, it would do it first go and every iteration would be identical.

I don't think so. I tried recently to prompt a badly sung and played instrument and it couldn't do it. I wanted some human error, something which sounded like 'instinct' but it couldn't. It only understands whats it's been trained on.

So feed it a reference track with the kind of sloppy playing you were after. Its training isn't limited, you can help extend it using reference tracks.

The AI can't replicate emotions; it doesn't UNDERSTAND emotions

Does your DAW understand the emotions generally associated with minor keys? No, it doesn't, but that doesn't stop it from allowing you to choose minor keys in a chord tool, does it? The human at the computer keyboard understands emotions and that's what matters. It's a collaborative process.

all it knows how to do is regurgitate its contents to satisfy the request as best as it can

Exactly. IT DOES WHAT YOU ASK IT TO DO. If you ask Gemini or Co-Pilot or ChatGPT to write an angry letter for you, it will write an angry letter. Try it and see. In the face of that, only a complete f**king moron would suggest AI can't understand human emotions.

So? The AI can write for me better than I can.

Almost certainly. Again, try it for yourself. Feed ChatGPT a set of your own lyrics, ask it to work on them for you and see what it comes up with. Get it to iterate a few times and see what you can cobble together. It can produce 100 lines of lyrics in less than a minute, from which you can pick and choose the bits you like.

If you really perceive that the AI can write better than you, then I can't argue with that.

That doesn't seem to be stopping you so far. But the thing is, it's not just writing better music than me, it's writing better music than most, if not all, of my favourite artists.

We only know what we know. It just doesn't feel very 'punk' of you to farm your feelings to a machine.

The medium is irrelevant, it's the message that matters. Nothing more Punk than that.

Yeah, so you'd be guessing, which is my point. It's hard to assess the value of something if i
I should really have released some of my other music. I don't only write insipid 70's love rock. Here's some badly recorded snips from a gig I did a decade or so ago

Not the worst shit I've ever heard but certainly something AI could do as well, if not better. There's nothing special about it.

How is your music not trying to evoke similar emotions to the punk-era of around 50 years ago but with an electronic ‘edge’?

I'm not trying to claim it's anything other than what it is. Any idiot could do it if they wanted to, none of it is actually difficult. A lot of you seem to think there is something special about what you do but that's just rubbish. 99% of it is krap that no-one is interested in, nor will anyone ever be interested in it.

And therein lies the pinchpoint. I actually enjoy just sitting down and noodling on a piano or guitar, now THAT's cathartic. You're looking to achieve, as you said, some kind of recognition of power within the people to be angry at (society? government? other regular shitty people? money systems? bus services? ..) the shit around them. And maybe to empower these people to make the changes which make them happy? Maybe that's the bit I feel is missing from your sound. It just remains angry; it's not looking for peace, which is what I do with my music.

Peace is easy, it comes with solitude. But I'm not looking for peace, I'm looking for anarchy. The end of civilisation cannot come quickly enough for me. 2040 is apparently the magic date so we won't have too long to wait.

ALL IT WANTS TO DO IS GIVE YOU THE NEXT MOST LIKELY <sample point/word (or actually letter)/adjustment to noise (when it comes to images> thing it thinks you want to see/read/hear.

That's fine, that gives us something to work with. What else does it need to be? It's people who try to anthropomorphise it. Corporations give it a human feeling front-end to make people more comfortable with it. That's not some grand conspiracy, it's just marketing. To me AI is nothing more than a tool, like a PC or a mouse or a MIDI controller. It just requires different skills.

[bermudagold]

Two recent productions of AI music I heard recently on YT were excellent from a musical and emotional aspect. The emotional intent was very human sounding and not artificial.

curious what you think of these two climbing the spotify charts

viewtopic.php?p=9187690#p9187690

[ghettosynth]

I literally can't see any similarities between choosing which AI generated audio from a single prompt you prefer to sitting with an experienced guitar player and trying different ideas (

You are 100% correct. And do you know what the key difference is? I can tell A.I. what to do over and over and over again. Again and again and again.
And it won't get pissed off, it won't be insulted that i didn't like it's first take, it doesn't need a break, and more importantly it will be available when I'm good and ready. Try that with some experienced guitar player.

It's dumb

Do the following equation in seconds: 2,673 + 4,654 + 5,212 x 3 + 8,789.
Ok times up. You failed. You're dumb.
But guess what? A calculator can do it.

The AI drummer also doesn't just suddenly fall off of his stool mid practice because it was at just that moment that the impact of that many beers hit and broke the frame of delusion that he could handle his drink. Yes, that's actually happened, I'm willing to bet that I'm not alone in that experience. For the curious, no, I was not the drummer in question. In fact, I'm not a drummer at all even though I am perfectly willing to pretend to be one if there is either a kit, or a star synare available.

[BONES]

No AI has been trained on my music, never will be.

Right, because your music is so f**king special, so utterly unique. Spare me days!

The thing is, though, that you can train it, you can feed it a reference track.

You have no clue about most music that's ever been made, that's abundantly clear, in the first place. Your view of things is about as narrow as can be.

Right again, because I walk around in a daze, oblivious to the world around me. Whatever helps you get to sleep at night.

We know you, you've not been obscure as to why you're in music. You have anger issues and you want to "express" that.

You don't know me at all. If you did, you'd know that I'm the happiest person on the f**king planet, for the simple reason that I am quite good at getting the anger out, at letting it all go. That and the fact that I don't really give a f**k about anything.

Hate is much more my thing, not anger.

Before you encountered this in music you weren't interested. You aren't interested in music, you understand fuckall about it.

That is no longer a necessary thing, no barrier at all to making music. e.g. I don't need to know what a chord is called to play it. Knowing it's name is only useful if you have tell someone else what to play, which is not something I have ever needed to do and not a position I am ever likely to be in.

I have my own ideas. AI has no ideas.

Yes, that's how it works - you take your ideas and tell the AI how to realise them. Just as you take the notes in your head and click them into the piano roll. Do you understand? The ideas are yours, the AI is just there to facilitate turning them into a song. Sure, there are plenty of people who will just say "write me a rock song" and be happy with whatever it spits out but refusing to use AI because of that is akin to refusing to work in a particular DAW because it includes a chord tool.

This is so basic, honestly, to fail to grasp it means literally unbelievable ignorance as to the act of inventing music

Right, because I've never written a song in my f**king life, have I? Or is it just that I have never written a song using methods that you approve of? Or maybe I have and you are making unsupportable assumptions based on your own biases? Perhaps we'll never know.

But here's an example that feels relevant. If you listen to Killing Joke, which I do a lot, you'll mostly hear single, sustained notes from their keyboard player, Jaz Coleman, or very simple, one finger riffs. Its a feature of their music and a lesson I often apply to the music I make. You'd listen to Killing Joke and think that Jaz was a pretty ordinary musician. But then you might go and look him up on Wikipedia and discover that he has been the Composer in Residence for New Zealand and for the EU. You can't always judge a book by its cover. Which is not to suggest that I am any kind of musician. I am not, I am a performer. But I am also a song writer and a producer, although those things are unimportant to me, probably because I find them relatively easy things to do so they don't feel like they are worth anything.

I am able to make my ideas happen a number of ways without much of a struggle. I don't, and won't know if an AI is able to somewhat ape things it finds some prevalence of trained on my music as it would be a massive waste of time.

Ignorance is bliss, eh?

My music is not reducible like that.

Of course it is. Music is maths and patterns, we all know that, and computers are really good at maths and patterns. To think that you might be an exception speaks of abundant conceit (something I was once accused of by a general).

This thing may be so dissimilar to the last thing, stylistically, in terms of materials... the instrumentation inside of one piece might represent two wholly different bands (in one case four)...

One of the random things Tunee spat out a few months ago was a bit like that. It started as a fairly standard, slower kind of semi spoken word EBM song then, after the first chorus, it injected a second chorus that was almost like a rock song - different vocal, different instrumentation - and then went back to the EBM thing. The thing is, it kinda worked. It's not one of the songs we're going to use but it was an interesting piece of music.

One thing might use conventional means, another is entirely off the wall. Some of those sounds never appear again.
in one case the microtonal aspect is not going to happen again, it stands alone to do this one thing one time.

You're not limited to using the whole thing the AI spits out, you can pick and choose, assembling a piece over time, using multiple prompts. You have to stop thinking of it as this thing that spits out 1,000 songs a day, it doesn't have to be that.

AI might get confused. There is no simple emotion, I do music. I did a show once where because of the political content or point - at the War Memorial Opera House - I did mean to tug on the old heartsprings by the end (I can sang country) but we did other things on that journey.

I don't know that AI is capable of being "confused", I don't think it's wired like that. But if you want to do something very complex, you can assemble it piece by piece, prompt by prompt.

Music to me, to a serious musician is not reducible to a need to express something.

Me, either. For me it is about evoking a response in the listener. If people get up and leave when we start playing, I'm fine with that but people who just ignore you and keep talking to their friends during a gig really shit me.

If you have a problem with someone 'clicking in notes in the piano roll', great let's have more of how you have no clue what music that isn't this coarse, primitive, simple-minded thing which expresses the same dull thing every time is, let alone can be.

More conceit here, that "simple" and "primitive" can't possibly hope to compete with your unlistenable garbage. Seriously, I'm listening to No Location at the moment and it barely qualifies as music. It's just a random selection of sounds, like the musical equivalent of a collage artist. It honestly sounds like it's all made up from random samples thrown together haphazardly. There really isn't anything to appreciate about it at all. Honestly, it's just plain awful. I'm sure it has an audience but that audience would be composed of the kinds of pretentious wankers l hope never to have to talk to. Which is a shame because, up until this point, I had some respect for your opinions on a lot of things. Having listened to that now, though, I am going to find it very hard to take you seriously in the future. Sorry.

You lack the very first clue as to what "AI" is or does.

No sweetie, that's you. I'm the guy who is actually working with it every day, both in my work and our music.

It has no emotional history, it has no thought, it has no capacity whatever to reason

Neither does my DAW or any of the other equipment I use but that doesn't stop me from getting things done. Yours is a nonsense argument. I'm not expecting the AI to provide the emotion or the thought or to reason, that's still my job. It just has to know what I mean when I feed it a prompt so it can give me what I'm after. It doesn't have to understand anything beyond that.

Also, to be clear, it's not less work than doing things in other ways. It's not the lazy alternative. I don't think I have ever worked as hard on an album as I have been working on this one and I can guarantee that my bandmate has spent an order of magnitude more time on it than any previous album. But the thing is, we'll get it done in half the time and we're enjoying the process a lot more, so it's worth the effort.

[BONES]

I will not use this shite to generate 'original music.' Ever.

You're not even willing to try it, to see if there is any basis in your clearly very strong feelings about it? Wouldn't it be better to come to that decision from a position of knowledge and experience?

I will not listen to anything that has anything to do with generative AI.
There's enough real music out there, to my taste, from Bach to Pink Floyd, to keep me sated.

What if you're unable to tell the difference? Maybe you won't even know? Honestly, though, I don't know what anyone hears in any of Pink Floyd's music. It encapsulates everything that made me so totally indifferent to music in my youth.

curious what you think of these two climbing the spotify charts

The Let Babylon Burn song is a great example. Assuming they are legit, you just have to read the comments to see how it connects with people. So Hollow has 23 million views and more than 12,000 comments. It's over, people, time to accept the fact.

[bermudagold]

The Let Babylon Burn song is a great example. Assuming they are legit, you just have to read the comments to see how it connects with people. So Hollow has 23 million views and more than 12,000 comments. It's over, people, time to accept the fact.

That's why I posted the Let Babylon Burn example...cause It's also a good reference example for the conversation in that it's so obvious the artists the vocal model was trained on...so if you were one of those artist does that change the calculus?...put urself in their shoes...this is where it can really rub people the wrong way

[Michael L]

"It's over, people, time to accept the fact" is what authoritarians like Putin and Trump say every day, as they try to convince citizens to accept a nihilistic view of the world, in which humanity is an illusion and only raw power counts -- in this case the technological power to steal other's creative work in order to make simulacra they control, in order to reduce the opposing power of fully human artists.

By the way, the AI defintion of simulacra is "representations or copies of people or things, often implying something that lacks the original's substance, becoming a copy of a copy, or even replacing reality entirely."

[ghettosynth]

The Let Babylon Burn song is a great example. Assuming they are legit, you just have to read the comments to see how it connects with people. So Hollow has 23 million views and more than 12,000 comments. It's over, people, time to accept the fact.

That's why I posted the Let Babylon Burn example...cause It's also a good reference example for the conversation in that it's so obvious the artists the vocal model was trained on...so if you were one of those artist does that change the calculus?...put urself in their shoes...this is where it can really rub people the wrong way

In general, with broadly trained models, like Suno, there isn’t any individual artist in the output in a literal sense. People are just hearing a familiar region of style space and mapping it to names they already know.

When a genre has a small number of culturally salient exemplars, any sufficiently good model will sound “like someone” because that’s how humans label regions of style space. The model has learned a smooth, high-dimensional distribution over vocal and musical features, and the output is a point sampled from that distribution, not a projection of any particular training example.

In fact, you are overvaluing this similarity and your mind is doing the work of matching the artists. Much of that comes from the narrow stylistic tropes that form the genre. It's not dissimilar from the McGurk effect, abstractly. Your mind fills in structure that isn't actually there.

Moreover, we have mechanisms in law to protect copyright violation, so, if someone thinks that this is a violation, they are free to exercise those rights. If it isn't a copyright violation, then what are they upset about, exactly?

[ghettosynth]

"It's over, people, time to accept the fact" is what authoritarians like Putin and Trump say every day,

Oh stop it. This is just a modern variant of Godwin's law. It was a statement of the state of technology and capability. Don't listen if you don't want to, nobody cares.

[Michael L]

Oh stop it. This is just a modern variant of Godwin's law. It was a statement of the state of technology and capability. Don't listen if you don't want to, nobody cares.

Yes, sir

[BONES]

That's why I posted the Let Babylon Burn example...cause It's also a good reference example for the conversation in that it's so obvious the artists the vocal model was trained on...

Is it? Maybe to you but it sounds kinda generic to me and I bet if you asked 10 different people who it sounds like, you'd get 10 different answers. And both the songs I listened to sound more like Country than Reggae. But the thing is, there are tons of vocalists who copy others' styles, it's hardly an issue limited to AI. The other thing, of course, is that the vocal sounds quite different in different songs. It's hard to get AI to be consistent from song to song like that.

so if you were one of those artist does that change the calculus?...put urself in their shoes...this is where it can really rub people the wrong way

It would make me laugh out loud. Honestly, imitation being the sincerest form of flattery, who could be offended or upset about it?

"It's over, people, time to accept the fact" is what authoritarians like Putin and Trump say every day, as they try to convince citizens to accept a nihilistic view of the world, in which humanity is an illusion and only raw power counts

Those are just the facts of the matter, as we are seeing played out currently by the Orange Idiot in the US.

-- in this case the technological power to steal other's creative work in order to make simulacra they control, in order to reduce the opposing power of fully human artists.

What a load of twaddle. Picasso said it best - "good artists borrow, great artists steal". It's been going on for the whole of human history, why is it suddenly bad because it's an AI, rather than some grubby little nobody?

By the way, the AI defintion of simulacra is "representations or copies of people or things, often implying something that lacks the original's substance, becoming a copy of a copy, or even replacing reality entirely."

And the relevance of that is what, exactly? Nobody has used that word, so why bring up it's meaning?

When a genre has a small number of culturally salient exemplars, any sufficiently good model will sound “like someone” because that’s how humans label regions of style space.

A good example of this is how people characterised my own semi-spoken vocal style, back in the 1980s. People whose exposure to music was limited to commercial radio and TV would suggest I sounded like "that guy from the B-52s" (Fred Schneider) because that was the closest thing they could compare it to. People with a broader range might suggest it was like Stan Ridgway from Wall of Voodoo, which was closer but still way off the mark. These days people just call it Heavy Metal because of the delivery, even though I couldn't do Heavy Metal vocals to save my life.

[jancivil]

Two recent productions of AI music I heard recently on YT were excellent from a musical and emotional aspect. The first was a trained classical musician give 5 or 6 audio clips of classical music, and asked to determine if they were from real composers are AI generated. He got two wrong I think and thought he was hearing a human classical composer. And thought the real human composer clips were AI.

Suno creating classical music is not a surprise. The 2nd thing I head blew me away.

First, why are you trying to sell this? If the goal is shilling for Suno it will do exactly this post.

Why would anyone that hadn't already caught this disease, wanting what you're sayng to be the truth, care about what you assert was excellent from any standpoint?

Someone we equally don't know from Adam was fooled as to whether or not examples were done by composers vs LLM. And we're supposed to look at this and see "classically trained" as if to lend some authority to your story. Trained to do what? Proficiency on an instrument does not amount to understanding music necessarily. Trained classical musician might be more trained performng seal than anything. Believe me, I know. A tiny minority of people in conservatory are composers. Training in music theory has two tracks in that pedagogy: for composers/for everybody else. So that lends no credibility in itself. <has no clue> has vastly more prevalence than one might think.

We don't know what this music was; "classical music" isn't any help in this regard. One might suppose, from "Suno creating classical music is not a surprise" that it's a music containing enough commonality, that has a maximal degree of prevalance formally and in the melodic & harmonic content that Suno is prepared to ape a generic style.

If that statement means "classical music" exactly or properly, this is a style period where generally there are few surprises. When there were, those composers are the ones whose names are known by everybody. Like anything else in the world, Sturgeon's law applies: 90% of everything is crap. So the metric <was it a composer or a well-trained LLM> speaks in no way to quality. Regardless o the type of music, most people suck.
So it's no surprise to me at all to see "Suno creating classical music..." per se. I can guarantee Suno is not trying to come up with something where they're confident of the test, 'was this Stravinsky or "AI"?' Stravinsky had several style periods that don't really resemble one another.

No composer has any use for this, the idea is you have ideas of your own and work them out.
The person that had use for Suno "creating" classical music has an agenda. I can't imagine any point in it other than someone marketing the product. Or someone being deceitful who wants to be recognized for these efforts but can't be arsed to create it themselves.

"Suno creating" is a corruption of language. It does no such thing.

This whole thing degrades the world.

[ghettosynth]

Yes, I think that you can tell pretty quickly. I gave Suno instructions to write a Bach style piano sonata, and this is the result. I can hear that this is AI after only a few seconds in. YMMV and all that.