Hacker News Reader

Comment by mcculley | Hacker News Reader

When models are capable of generating music better than humans can, Spotify might be disintermediated entirely.

The next logical step after AI-slop music from Spotify would be for device manufacturers to cut out one more middleman, and provide their own locally-generated on-device AI-slop music as a feature of their devices.

Why would slop be the only outcome?

Even if the AI produces a pearl, it is going to drown it in shit. Because it is an automated operation.

No new AI content generation advance has so far escaped being used for the equivalent of SEO spamming, long-tail flooding the market to mislead consumers and lazy “passive income” bullshit.

No AI content generation advance ever will; it’s not like industry is going to hold back and say: not this one, guys. It’s too good. We’re going to let it have three years in a custom built home studio to make its magnum opus.

I think you are conflating how models are used by services like Spotify with what they can do. I run some models locally. These are going to become more powerful.

Your confident prediction of the future limitations of models sounds like what the chess and go masters said.

I'm not conflating anything. I am just answering the question of why slop can be the only outcome.

On a Spotify-level scale, web-level scale, slop can be the only outcome.

The question I asked was specifically about locally generated content.

But it's in the context of an alternative to a Spotify product, right? With millions of users.

Exactly where the slop is generated, or by whom, is essentially irrelevant: at scale, slop is the only outcome from these tools. It always will be.

And presumably in a resource-limited environment, slop quality is even more likely.

Billions of users, with increasingly powerfully hardware, locally executing models that better understand the world and human psychology.

The alarming thing in this little sub thread is that I am not wholly certain that none of you are joking. That is where we are now.

I am not joking. I am also uncomfortable accepting that humans will eventually not be the best creators of art enjoyed by humans.

Then don’t accept it. Because it is fundamentally untrue.

Why do you think it is anything other than inevitable? You think that models cannot learn how to make music that humans will enjoy?

You are watering down “the best” here, IMO. This definition leads towards “the most efficient production”, not “the highest art”.

But sure, if you don’t care that a human isn’t involved in doing it. Personally it’s not negotiable to me and I think more people believe this than you might expect.

I do see it as inevitable that models will create the best of everything. I completely understand that some humans will choose to reject anything not created by a human.

Based on what evidence? AI models haven’t created the best of anything yet.

Actually AI hasn’t even made anything I’d call very good thus far in my personal experience.

I agree that current models are not as good as humans. They will continue to improve. They will eventually be superhuman, just as they are at chess and go.

As the guy who started this subthread, I was joking — but I expect it will come true anyway. The whole industry seems bent on removing humans from the production of all goods and services, so someone somewhere will try to optimize humans out of music and the humanities in general. Whether they succeed is an open question.

Isn’t every industry trying to eliminate human labor? Nothing will prevent humans from making music. Being paid for making recordings was already a dying business model due to digital reproduction making copyright toothless.

The music industry is at least 10 or 20 times older than the concept of copyright. The traditional business model was always about live performance and patronage, and even today that’s where most musicians’ income comes from. The recordings and radio/streaming play could make money, but it was mostly advertisement for the shows and merch. So functionally, the advertising machine is shifting towards making culture a mere commodity, which is a new thing.

The other thing here is that while it is a continuation of industrialization of artisan work like furniture and textiles, we are finally going into the area of commoditizing the work of the artists. Admittedly the split between artisans and artists only happened recently in the Renaissance, but it’s still a different threshold, where culture itself is mechanized. And instead of asking if it is a good idea to take the humanities out of the hands of humans, the only real consideration will be short-term shareholder value.

I am, of course, well aware that musicians and the music industry existed before copyright. I am questioning the possibility that humans might be successfully optimized out of music or even entirely out of the music industry. Even when models are better than humans, humans will be free to make music, just as chess players are still free to play chess now that humans are no longer the best at chess.

The point about algorithms driving culture is more interesting.

Reply on news.ycombinator.com