Spotify: Droppin' Some Fake Beats | Hacker News Reader

Spotify: Droppin' Some Fake Beats

lcamtuf.substack.com

101 pointszdw81 comments

The post and comments come out a bit unfair if you read the answer from the music author.

The music author seems to be doing the music as part of a thesis and does not dump the generated music straight from the model. Instead it uses it for ideas and cooperation, and in the end there is post production and real work. I know people doing a similar thing to learn how to write books with AI as a productive tool.

In my opinion this is the real promise of AI: elevating the productivity and enabling more people. Like the internet has done, it will give voice and tools to many who could not, and a lot of those “voices” may seem of low quality, but alas such is the price.

As in every technological progression, paradigm changes happens. Some good some bad. Give yourself some agency and avoid the bad content and enjoy the good ones.

Also if you look at the YouTube videos there are comments of people genuinely liking it.

Finally some refreshing view: I am constantly bombarded by people who either completely abhor or love it.

I think it's a tool, it's going to take away jobs which were already shaky and on the line, while empower people who can understand its potential and limitations.

Yeah, but that kind of thinking isn’t good for attracting media attention or VC dollars.

The headache for me is triaging through lots of quantity of grbage to find something of quality. The question to me is, do we really need all this new junk in the first place?

Still better than some rich guys choosing young people to have them take out a huge loan, feed them a bunch of lyrics some dudes wrote across the sea, match it with one of the 200 ready to go beats their "label" has in the drawer and make a single with a music video to pump tiktok ads on.

There's very little romantic songwriting and true exploration going on in mainstream music. AI productions if anything to me are more honest if done with an open model by a person in their home. At least it's just a person fiddling with it until it sounds as something they like and not a megacorp. Until spotify makes these themselves.

The fault in this logic is that exact same people will be leading the way of mass manufacturing more soulless garbage and dumping it to the consumer as mainstream music. AI doesn't actually fix the problem, it just makes it easier to do.

If you read my comment you see my last sentence is there to acknowledge this!

What do you mean "take out a huge loan"?

I assume they mean an advance. Advances given to artists upon signing with a label are paid back out of proceeds from sales/royalties.

So in other words, not like a loan?

"like" aka "similar to", aka "shares some features"

An advance is "like" a loan because in both cases, the money is expected to be paid back by the person or group of people who receives it. There's lots of differences. Repayment terms. Exit clauses. Interest rates etc. But seems pedentic to say that the advances offered to musicians are not like loans they take so that they may have a claim on the future revenues of the sales produced by their work.

Well yes, but flippantly describing the arrangement as "rich dudes convincing poor dudes to take out a huge loan" makes it sound far more exploitative than it actually is.

It’s astounding how often people conflate advances and loans when talking about the music industry. If anything it’s closer to seed capital than a loan.

Given the low ratio of “making it” to “wanting to make it” I would assume that there are many people who DO actually take out loans.

Joe Satriani famously self-funded his first album by taking on large amounts of credit card debt.

I'm sure there are many, many other musicians who took a similar risk that didn't pay off as well as it did for Satriani

Don't you have to pay back advances? Never heard of that happening for failed investments outside of fraud

Advances are recouped from the sales of your albums or tickets. Assuming you’ve made a good faith effort to fulfill your obligations you don’t need to repay the advance or pay interest on the advance if your album doesn’t sell.

> Assuming you’ve made a good faith effort to fulfill your obligations you don’t need to repay the advance or pay interest on the advance if your album doesn’t sell.

It depends on your contract. Usually, poor sales won't result in needing to repay an advance. However, there are plenty of other clauses that the contract could contain that would lead to the contract being canceled and potentially the advance having to be repaid (due to delays, editorial disagreements, poor quality, etc.)

Spoken like a non-creative

Mainstream music has been saturated with commercial influence for decades. We've had many, many broad revolts against this, that more or less immediately got picked up by the entertainment industry, commodified and commercialised.

Right, if we reductively summarize any process then nothing really matters. Its sort of like saying "hendrix operated a string instrument"

Spotify (like other streaming services) pays about 70% of its revenue to the rights holders of the music it plays.

AI music means near-zero cost of goods sold so the full 70% can turn into profit. The incentive is so strong that it's a certainty waiting to happen.

I hope I am not wrong, but there is so much music out there already that success today is not about the quality of the music but instead about the audience's connection with the artist.

If there is no artist I don't see how any strong following could be built. (Obviously AI will carve easily into elevator music etc.)

Who will ever look up a song and favourite it on Spotify if there are no humans behind it?

People will just follow curators, who pull the right AI music into the right playlists.

People like me who mostly take songs at face value, valuing them by music and lyrics, and don't really care about who wrote or performed them, much less what else they're up to in their lives?

That said, the quantity of music is in itself a problem. Half the value of art is in experiencing the same thing as others. It's going to get even harder with cheap generative art. The better it is, the worse the situation gets.

Personally 3/4 times I'll put my big Spotify playlist on random or play a Spotify curated playlist. Certainly there is scope there for Spotify to select songs based on their licensing costs.

I've certainly had a hunch that certain songs get much more frequent rotation than others on my liked songs playlist (although I appreciate humans are notorious for picking out patterns where there are none!)

The "relaxing piano for studying" type playlists get a huge amount of streams, and they're among the easiest to fill with AI slop

I haven’t cared about the artists (their biography or something?) since I was a teenager. Just the music.

But I’m also the kind of person that hardly ever goes to concerts.

> Who will ever look up a song and favourite it on Spotify if there are no humans behind it?

"Obscurest Vinyl" has 249,608 monthly listeners on Spotify. No one is interested in the humans behind their hit "I Glued My Balls To My Butthole Again".

K/DA and Hatsune Miku would like a word, perhaps.

I view them (and Gorillaz) as counter-examples.

I feel like more people are familiar with Miku than her music.

I rather think that makes them examples on point, specifically on the point that:

> success today is not about the quality of the music but instead about the audience's connection with the artist

...although I would rather say "sense of connection", since in all three of these cases there is in the narrow sense no "artist" involved at all.

More broadly, this seems a fairly uncontroversial point, given how crowded by supply our current media markets are made. Certainly it seems like it would be uncontroversial on that supply side, at least if what I hear from artists and writers and musicians in my circles is any guide.

Like 10% of the music I listen to is a deliberate listening experience: I sit on the sofa and play something that I like.

The other 90%? Background music, e.g. for daily exercise, where I mostly want upbeat music that keeps me moving and sounds good.

For the 10% I prefer humans, for the 90%, if the AI music gets good enough, I'm afraid it will do the job just fine.

As an anecdotal statement, I do not want a connection or care about the artist. I also don't connect (or want to) with directors, writers, actors etc in TV shows and movies I view.

Music is just another consumable form of entertainment for me, and if I like the song I don't care who made it.

A while back there was an article circling about how Spotify has their own contract musicians and how they are pushing "fake" bands made up of these musicians in their playlists. Same reason. With AI they don't even need those musicians.

Most musicians I know would love one of these Spotify jobs. Writing and recording for a salary, rather than spending 75% your time on (mostly futile) promotion for a wildly unpredictable return?

With Spotify earning only 30% what’s their incentive to not push for generating and getting 100% if the people will listen.

(Keeping in mind actual artists see a small amount of that 70% anyway so it’s more an argument about paying labels)

Maybe this can finally make record labels get their shit together and provide reasonable licensing for music.

Ever wonder why restaurants and other commercial venues keep playing bland bossa-nova cover versions of good songs?

Licensing.

I recently learned that Spotify has a service for this as well https://www.soundtrackyourbrand.com/

> Despite some HN quips, AI text detectors are pretty dependable

I have never tried AI text detectors, but my impression was that they were considered unreliable?

They are. The author is wrong.

If you ever dealt with horrors of "plagiarism detection" tools...

It only gets worse.

I created some AI Audios based on ChatGPT/ Suno / Midjourney / Joined via ChatGpt

This one even asks the question, in german: "Is this art or can I throw it away?" (DE: Ist das Kunst oder kann das weg?")

Can AI music trigger real emotions or are they just artificial emotions?

https://www.instagram.com/reel/C79y1Dbst3l/?igsh=bDgzeGMxZTJ...

(And this one I call "If Poe had had a cat." https://www.instagram.com/reel/C79yC-rsywz/?igsh=MThzbDZvamV...

Title by me, everything else Chatgpt, Suno, Midjourney and ChatGPT again.)

So the big question is, can AI generate art? Can AI generate "real" music? Or is all AI music artificial music? Can artificial music create real emotions?

I love this topic. So much possibilities to challenge our understanding / appreciation of art and music.

It seems obvious, at least to me, that AI i.e: code written to output something trained on human input, can make "art". It doesn't seem like a particularly deep or meaningful question. The question for me is whether you choose to attribute any value to it. I can hear a song that AI made and note that it sounds good (although this is rare currently). I have no interest in investing time into it though. I'm interested predominantly in the human experience and what humans express through art. Art created through the push of a button (or prompt) is still art, in the same way that pieces created by Warhol's assistants is still art. It's just indirect. That indirection, which for AI is on the extreme end of indirect, make it a bit of a novelty.

> I can hear a song that AI made and note that it sounds good (although this is rare currently). I have no interest in investing time into it though.

This argument makes sense to me for e.g. a novel, and is the reason why I wouldn't read an AI-written novel, all else being equal with human-written novels. But much of the music people consume is background music and requires no time investment at all.

For example, I play music while I exercise... give me a playlist that sounds good enough and it's upbeat enough, and it will be fine for the task.

That 2nd one is actually really good. :)

I agree, the rhythm of the rhyming feels great - tight and (dare I say it) artistic.

@franze is that something you have control over - is the prompting so detailed / accurate that you can specify that? Or is there some other tool that can adjust an initial revision? Or is it simply a roll of the dice?

A lot of the credit here is owed to Edgar Allen Poe for writing lyrics that are so damn catchy.

But yeah, Suno managed to do a decent job setting it to a poppy beat that is well-suited to the poem

A lot goes into the lyrics. And yes, I found using existing (public domain) poems as a basis greatly improves the ChatGPT created lyrics.

With suno there is a lot of difference between the models (v3 vs v3.5) and then it is a lot of sometimes pretty random style prompts and a lot and lots roll of dices.

It can get quite frustrating, i.e.: https://www.instagram.com/p/C7oeyGzt0jd/ its unfinished and I like it a lot - but I never got Suno to recreate this exact style again.

Yeah, that one's all over the place but has some fun, bright spots that are really interesting.

Can see why you like it.

Trying out that Suno thing now. Seems super easy to get workable results.

ie: https://suno.com/song/dfaa6a3c-ce91-4b71-a394-5b9232d00dc7

I can see why legit artists are concerned. (!)

Suno and friends, IMO, make good to acceptable songs. A few times per genre (for those they even can do, growling is still outside their capabilities). But, as with most things AI, it becomes very recognizable and same-y after a short time.

> it becomes very recognizable and same-y after a short time.

Yeah, I'm starting to get that too already. :(

v3 gives more variation, v3.5 is all very Taylor Swift style. seems totally overfitted.

Ahhh. I'll try v3 out tomorrow. Out of free credits for today. :D

If you wanna merge the audio with some pics to easy shareable videos, you can use this tool: https://image-audio-video-merger.franzai.com/ coded by ChatGPT of course.

Awesome! You should add some kind of "automatic import from Suno" for the audio. :)

I know a paper on this topic [1] whose take I really like: that AI generated works only become art through the eye of the artist.

In isolation a sunset is physics, photography is chemistry and AI is math. Without the artist there's no real difference between an AI song and playing /dev/random. Only through human interpretation do astronomy, chemistry and math turn into art.

[1] https://journals.sagepub.com/doi/10.1177/13678779241252664

I really love Udio, it works more like a tool for artists to extend, inpaint and modify musical ideas and explore. Compared to Suno it feels less polished but the AI is slightly more "dense" if that means anything anyway...

So long as I know it uses AI generated vocals or instruments it's A-OK with me. It's an emerging genre. Whether it's junk is going to depend a lot on how the director corrals the output. I've listened to plenty of junk already. Amateur work can be fun to try and I'm curious how this will compare to Vocaloid or autotuned stuff.

  >So long as I know it uses AI generated vocals or instruments it's A-OK with me.

Can you not tell the difference?

If you can, do. If you can't, why?

Curtis Roads, in the latest edition of Computer Music Tutorial:

> A new industry has emerged around AI services for creating generic popular music [...] This is the latest incarnation of a trend that started in the 1920s called Muzak, to provide licensed background music in elevators, business and dental offices, hotels, shopping malls, supermarkets, and restaurants.

Muzak aimed to be more than just background noise - it supposedly improved worker productivity, and manipulated your mood.

I don't understand how a real generated song is classified as fake? This just doesn't sit right with me. It's as if artistic expression is not valid if not done throught traditional means. This is the latest fad, to look down upon people who are exploring newer tools. If we don't accept experimental things like this, it will be relegated to only the spammy type of people.

Oh but wait, it gets worse. I hear some of these new "artists" these days are using "synthesizers" and artificial "acrylic" paints in stead of banging rocks together and smearing blood on the walls of caves. Artistic expression is dead!

Take a picture, it will last longer.

Congratulations!

You just destroyed painters' jobs.

What's next? Take a man's horse! why dont ya?

The "before:" operator trick is very useful but it won't last. They have quietly phased out some of these operators and they are not supported on many google properties. The before operator is also buggy, I regularly get searches where they fuzz the date by years.

Are machine generated podcasts also a thing? My gut feeling is that it would've harder to make passable AI podcasts but I can't give a rational reason why. Spotify would be all over that.

Of course. Podcasts were the first to be "disrupted" and flooded. I wouldn't be surprised if most podcast releases are AI generated at this point.

Do you have examples?

>A while back, I made a rather fatalistic prediction: that barring a radical intervention, in a decade or two, most of exchanges on the internet will be faked. It’s not pessimism; it’s that our aggregate capacity for human-to-human interactions is inherently capped. In contrast, the ability to generate human-like text, images, and audio is now almost infinitely scalable — and from customer support, to marketing, to cybercrime, there are powerful incentives to crank it up to eleven.

I'm waiting for the first ChatBot to talk to customer support ChatBots for me.

I think artists should get some option of marking the music as AI generated. Right now that option does not exist on Spotify even if you want to.

I wonder if long term people will actually care about that.

And of course I want robots to get off my lawn, but I don’t know whether younger or upcoming generations will share that sentiment.

In the past there was similar reserve about music produced with autotune, samples, not recorded in a single take, etc.

It’s more about not training AI on the output of other AI’s.

Hopefully the enshitification of the net kills the social net splintering the web into a pre facebook era.

This is an impossible dream. Facebook has existed and the arrow of time goes only one way; there is no longer any possibility of a "pre-Facebook era," because that era has come and gone. What we would see instead would be a post-Facebook era, wherein we would still have to reckon with the social damage it and its ilk have inflicted on a fairly significant fraction of the species.

It isn't a new sort of error, if that's any consolation, being in the limit no younger than the story of the Garden of Eden. But while studying the past can prove valuable in approaching the future, hoping to return to it never does.

Controversial take ahead. If, as an artist, you don't want AI to eat your lunch, there's a simple solution: produce art that is better than AI.

It doesn't seem surprising that the AI band here is in the metal subgenre. Some metal is great, but there are also an awful lot of mediocre, repetitive tunes and ridiculous lyrics. It's not surprising that it's easy to imitate! It's not very good! "Old Gods of Atlantis"? AI or not, that's not a promising song title.

Show me some actual good art that AI has produced, then I'll worry for you.

> Show me some actual good art that AI has produced, then I'll worry for you.

Does audio count? If so, how about the 2nd link here:

https://news.ycombinator.com/item?id=40644102

No, I think that is pretty bad. I mean, it's awesome that a computer can do it, and the concept of tweaking the Poe story is fine - but I suspect that is a choice made by the prompt engineer. As for the lyrics, throwing in the ember after remember/December/ember is a tolerably cute rhyme, but "my heart was aching, young and breaking" etc. is just standard, and the tune is absolutely generic. There's nothing new or interesting here except the fact that it was done by AI!

Ahhh well, I liked it.

My standards may not be at all high though. ;)

What do you reckon of the first song ("dialectic (accept it) - Yolkhead Radio") here?

https://suno.com/radio/song/e48c0ee8-24c8-459b-8d91-469dbd6f...

Reply on news.ycombinator.com