Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.
- 223
- 3
Jump in the discussion.
No email address required.
Notes -
Suno, the AI song generator, just released version 5. And now I think we are 100% past the uncanny audio valley. Version 4/4.5 was at the level of "Convincing, occasionally incredible, but still flawed enough to notice." Version 5 is 'tricking' my ear 9 times out of 10. Studio quality. We have a fully functional infinite music machine available for the monthly cost of a cheeseburger.
A few examples:
https://suno.com/s/N86w28eQjBWbI6fA
https://suno.com/s/BsKe5OnQpUhPj2Zx
https://suno.com/s/voPPxtsXxRjFRF93
https://suno.com/s/Yqe3pzUQHIPAQ4g4
I think people get too focused on the apparent 'slowing' of progress in the LLM space and think its proof that Machine Learning itself is not living up to the hype.
Meanwhile stuff like Video generation, Music, and Protein Folding/Drug Discovery are still improving rapidly.
Arguably LLMs are just the interface by which we can access these other powerful Djinns to provide us with the particular services we want, as we await the "one true superintelligence" that can do anything to arise.
There's probably a small window right now to write a Sci-Fi novel that features humanity invoking individual AI patrons that specialize in particular aspects of the world, in the same vein as 'old gods' (Stockfish God of Chess, Suno God of Music, Midjourney God of Aesthetics).
Anyway, if there was ONE arena you would want AI to reach superhuman capability, one particular application that would improve your life even if AI progress stalled out otherwise, what would that be?
For our purposes, lets just grant "customized pornography" as the killer app.
Me, I think I want the ability to produce bespoke episodes of older TV shows that I enjoyed but were cancelled or went off the rails and/or had horrible conclusions. GoT and Firefly are obvious examples there. But I have several others in mind.
It'd be cool to live in a world where the "Canon" of a given series was not defined by any particular "official" source, but instead you had a whole library of 'forks' in the plot and character development that fans can choose from, or generate their own as they like, with maybe some curation done by the rights holder to identify the entries they deem 'high quality' and consistent with the original vision.
You don't notice the very clear artifacts in the sound?
I don't really understand why these exist and would have assumed that would have been easy to fix but seeing as they haven't been perhaps it's much harder than I imagine.
Beyond that the music itself is bland and the audiomixing poor, but perhaps these are prompting issues with these particular tracks.
If the technical issues can be ironed out this seems fine for stuff like background music in shops and receptions. Given that a Spotify subscription costs as much why would anyone switch though?
What is the use case here? I'm not being facetious, I think this is cool as hell but what is the use case? Composing and recording music is already dirt cheap. Extremely quick prototyping? Perhaps as a way to create temp music?
You get 3000 credits a month for $10, and each song gen costs 10 credits. So you can get a couple hundred songs (assuming you're putting in a little bit of effort to tweak each one) each month for less than the cost of a CD back in the day. Whether that's a sustainable price point I don't know.
So we're talking below dirt cheap at this point.
MY use case has been making new high-energy songs to slip into the playlist at the gym, which is a fun process.
But this is also a step towards what I suggest is my preferred use-case: to make bespoke TV episodes, it'd need some ability to compose soundtracks and theme songs and such.
I've been using Suno for about a year and a half, and its gone from "Constant artifacts that instantly betray its AI" to "If it played on the radio I wouldn't peg it unless I was paying close attention, and even then I wouldn't be sure." It even adds in respiratory sounds for Pete's sake!
Or more to the point, I think that if we did a double-blind test with randomly chosen people listening to AI songs vs. decently skilled indie artists, 80+% of them wouldn't reliably catch which were AI and which weren't, if we curated the AI stuff just a bit.
Funny you should mention that.
Owing to the absolute dirt-cheapness mentioned above, its a 'viable' (if you cheat) business model to mass produce barely passable songs and upload them en masse to every streaming service under the sun.
And its actually debatable if this really makes the services worse given the fact that most users don't seem to notice or care much.
It certainly makes it harder for new, undiscovered artists to stand out. And that's the one thing AI has going against it. You can't yet attend a concert for an artist who only exists digitally.
But mark my words now, the first large music festival showcasing ONLY AI-produced music will be happening inside of 5 years.
Also, like two years back I talked about how I was still collecting music to my local devices through force of habit. It seems even more laughably futile now in the face of tech which can keep producing songs faster than I can even listen to them.
And more recently we discussed the art of cover songs.
This is also a machine that, if legal restrictions were not an object, lets you translate any given song to any given genre, instantly.
Ahem...cultured people have been doing this for a while now.
Yeah, its a very solid point.
THAT SAID... I'm wondering if people would actually be willing to buy tickets to sit and have the AI's songs just played at them over the speakers, even if there was an video accompaniment.
Or they can commission an actual band to play the songs, but at that point... just become a fan of said band?
Well, guess we'll see if I'm right:
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link