General AI thread

[BomberPat]

34 minutes ago, Hannibal Scorch said:

Oh and if you do use it, remember to say please and thank you. They will be our overlords one day after all.

There was a news story the other day about how saying please and thank you costs OpenAI millions of dollars in additional energy costs, presumably because it prompts an additional unnecessary response.

12 minutes ago, Sphinx said:

It's learning from patterns in data and creating something through prompts to sound like something.

Human music isn't immune to it. Like Paul McCartney imitating Little Richard or Oasis ripping from The Beatles or whichever indie band ripping from them.

It isn't "learning", and there's no comparison between pattern recognition by a machine and human inspiration or taking influence from previous artists.

AI doesn't learn, it collates data and extrapolated from it - if you ask it to draw a picture of a dog, it analyses everything in its database marked "dog" and extrapolates from the shared characteristics of those images. It's one of the reason it struggles/struggled with hands - it has no understanding of the structure or purpose of a hand, it just "looks" at everything that's been labelled "hand", but that is everything from open palms to clenched fists, so it tries to find the common features and comes out with a mess of fingers.

When it comes to music, yes, bands can knowingly or unknowingly rip each other off, or they can use samples, or reference riffs and motifs from previous songs, but they do that knowing the context of those previous songs, knowing what they hope to achieve from incorporating them in their own music, and are capable of crafting that into something new. Venetian Snares released a version of Gloomy Sunday that samples Billie Holliday's version, but transformed it into something completely different - in theory AI could do that, but only if you think of music as nothing but an arrangement of sounds, rather than something that exists in a broader context between artist and audience - that Venetian Snares track is powerful because of how it distorts the original, which relies on the listener's familiarity with that song, and the stories behind it, for full effect. 

 

Ultimately, I don't know why anyone would even want AI generated art, music, film, or literature. What is it for? What is the point in art that tells you nothing about the people who made it, or the context it was created in, and has no intention at all behind it? And it completely overlooks that creating art is never a matter of just having an idea and then seeing it to fruition, it's about everything that happens during the creation. AI could never make Jaws, because it could never account for the mechanical failure of the shark and the need to rewrite the film around that, and how that actually massively improved the film. AI couldn't make Star Wars, because Star Wars barely exists in George Lucas' original script, it was born in collaboration with set designers, concept artists, actors, and everyone else involved, and even then, was reshaped by accidents, by weather conditions forcing their hands when shooting, and all of those random events and improvisations that actually shape art, which AI fundamentally cannot replicate and will never be able to replicate. The same is true in every other medium. 

 

Ultimately, I just get annoyed at the idea that AI will let you create whatever you want. Because, for a start, it won't, it will just be able to rearrange existing data and steal other people's work - and it's telling that its use cases so far have been actively demonstrating how well it replicates other people's work, rather than producing anything new. But more than that, I don't want art to be exactly what I want. I want something I never knew I wanted, I want something new. And AI, by design, cannot give me that and never will.

[organizedkaos]

Sampling is a great example too. Someone once described samples in hip hop as "the wall the graffiti is on". It's not just taking a bit of music and making other music from it, it was a wildly creative thought, to take existing, often mainstream, music and use that almost as a rhythm for your poetry.

Sure AI can piss out hours of chilled hip hop beats to study to, but that's just music as ambience, or a distraction.  It's just some drum beat to move your oars to. It won't ever do anything creative (beyond some vague sort of outsider art stuff, the creativity of which only comes from our perception of it)

And I work in tech. Whilst there's plenty interest in how it can be used (unsurprisingly it's really useful for mapping out stuff in programming languages) noone has expectations beyond its capabilities. Not thinking much of it is not so much being against change, it's an awareness of limitations.  In fact that the reason there's so much excitement about generative AI is almost the reason it's limited. Human beings are capable of fantasy, creativity, empath, narrative and lateral thinking. We can't not perceive AI through lenses that make it a potential new conscious being, it's possibilities seem limitless. We've started thinking of the furby as a real pet. The furby on the other hand still just wants to tell you the next most likely word in the sequence

[SuperBacon]

1 hour ago, organizedkaos said:

Sampling is a great example too. Someone once described samples in hip hop as "the wall the graffiti is on". It's not just taking a bit of music and making other music from it, it was a wildly creative thought, to take existing, often mainstream, music and use that almost as a rhythm for your poetry.

Great point and to go slightly off tangent, sampling is often seen as just lifting whole pieces of music and repurposing it, when if you go into the detail and see how someone like Premier lifts, chops, shifts, moves notes around from their original status and makes some of the best music you've ever heard, you start to appreciate just how amazing it is.

Sometimes I'll listen to a sample and think how on earth has his mind worked to see that as the end product he came up with. It's incredible.

[Sphinx]

I think art can be rewarding and nourishing without knowing about the people who made it or how it came to be. When I hear a song for the first time, I don't necessarily know the artist's backstory or anything about the artist's identity. I could be in a club hearing an electronic music track with no idea of how the song came to be - other than its context next to other tracks that were played that night. But I can love it and know it moves me regardless. I'll love it more if there's an interesting story behind it, but it's not a requirement. The music itself and the context of where I'm hearing it, what I'm hearing it through, and where I'm at in my life contribute to the emotions I feel.

If I dug down into that track I heard, I'd find there's certain sequences of notes that are played, sequences of chords, the dynamics of the notes played, which instrument played it etc that gave me an emotion for some reason. The same would apply if I were to analyse a classical composition.

I think that AI (or Machine Learning?) could look at that mass of data across every piece of recorded music on the internet, find patterns as to why we humans like certain compositions and what stirs emotions (what's critically acclaimed, what's commercially successful, even looking at time stamps on YouTube video comments of people saying '1:58 insane solo' or whatever), and then draw from that information and a vast library of sounds and virtual instruments to create something that also stirs emotions and/or would be considered a good composition on a technical level in its application of music theory.

Data from our own specific listening habits could be fed into it. Say Spotify introduces an AI feature that generates compositions using data it's gathered through your playlists, what you've listened to and when, what time of day and month you've listened to it, when you skipped a track etc. It would be tailor made to you and, by sifting through that data, would understand patterns as to what is likely to resonate with you. I'm not saying that's my ideal - I'm saying that would give it a good amount of data to come up with something that'd be pleasurable to listen to for that specific person.

It could also consider the history of artists, history of the genre, what was happening in the world at the time of songs you like or that it's drawing inspiration from, trends, what the song sampled (if anything), what it seems to have been influenced by etc. and then use that to inform its creation. It's not like what you'd hear would necessarily sound like a Poundland version of a more popular artist - it's pulling from masses and masses of data from tons of artists.

On the point about wanting something you never knew you wanted, the world champion of Go played AI in five matches and won once - the only time AI has been defeated in a competitive match. The human said that AI was playing moves that looked as though it had made errors, before realising several moves down the line that it had actually pulled off moves that were genius and that a human would never think of. I don't know whether you'd say that what it achieved was creative, but it seems it's capable of recognising patterns that we might not understand or don't perceive and making unorthodox decisions to get a winning outcome.

I imagine I may lose people here if I haven't already - I know Go is strategy and art is art. But if AI were to 'strategise' how best to create a song within certain parameters, I think it's possible that one day it could generate something that would both surprise and - put next to a quality human creation - could be considered high quality or as emotionally compelling, if not more, as the human version by a large number of people. I think that would be even more the case if it was used in collaboration with humans - as a tool. A human singing vocals rather than an artificial voice for example, or humans playing the instruments for the composition that AI came up with. I also know that sounds depressing - art as a problem to be solved.

But yeah, in terms of art being human expression, of struggle, of identity, then no it doesn't fulfil that.

Edited April 24, 2025 by Sphinx
My font's bigger as I wrote it on my Notes app - does make it look like I got ChatGPT to write that though

[TibBo]

Yeh I think you are missing the point of why people create art/music etc.

If a songwriter is a songwriter, then chances are they enjoy writing songs and find some kind of fulfilment from it on a personal/emotional level beyond just critical recognition, so just doing karaoke over an AI generated track would be taking away part of the process that they are passionate about.

On the flip side of that, if you just want to sing songs for the solely recognition/adulation without putting any effort into creating the songs, you would just sing covers of songs people already know, and go the tribute act/cover-band route.  Anyone that has ever been in an original band playing to an empty room will understand that struggle compared to playing shoddy covers to a room full of people.

This would only solve a problem if you wanted to be a songwriter, but didn't want to/couldn't write any songs yourself?  In which case maybe a career change would be a better option?

[Your Fight Site]

15 hours ago, BomberPat said:

Ultimately, I don't know why anyone would even want AI generated art, music, film, or literature. What is it for?

This is my bugbear with “A.I. all the things” to be honest. LinkedIn is the worse for it.

If you have A.I. generated posts, and A.I. generated responses to those posts, then what’s the fucking point? You’ve just got bots talking to each for absolutely no tangible game. You’re not informing any one, you’re not educating any one, and you’re not entertaining any one. It’s just… pointless.

Also, all these companies laying people off en masse to replace them with A.I. workflows or whatever, are also going to find they’ll have no one to sell their products and services to if a large percentage of the population is unemployed because “A.I. took der jerbs”.

Edited April 24, 2025 by Your Fight Site

[organizedkaos]

I feel there's a heart then head with the best of music. As Sphinx said, often one doesn't know why something is working when you initially hear it and that's great - but then often the love grows (certainly for me) as I understand better what's going on in the song/film/art beyond simply "wow this got me". I'd also argue the connection often comes implicitly from the thought the creator has put into it. I do think there's a bit of an uncanny valley to AI generated music, it might sounds good but it feels empty (or maybe that's my bias from knowing it's AI).

Electronic/club music is actually a really interesting example to use here. Sure a lot of what DJ may play in a club is simply to get people dancing, and as pointed out you can feed data about that music to an AI and it'll probably find the connections. But the best club music is constantly evolving, it's in a conversation with itself. Say, underground techno in the last 15 years goes from being very of the underground resistance style to minimal to more house influence to heavier pounding techno again to heavier techno with more trance influences. This is because whilst many DJs chose whatever are the big songs of the moment the best are experimenting with new sounds, noticing what works on dancefloors that's outside of the paradigm and then pursuing that. You have people like Erol Alkan and Seth Troxler who have helped reshape what the current in sound is based on things they like and think the audiences might like, finding songs that display a certain sound to create essentially new movements and genres.

AI essentially just maintains the norm. It's like a better version of begginer student DJ just playing Beatport top 10....

That's before you get to someone like Midland who sure, creates "Final Credits" as the "sample an old funk song" style gets traction (and that song rules) but continued to evolve and change his style rather than resting on "this is the thing people want". Then he creates an absolutely beautiful album that uses clubby music to look at both his identity as a gay man and the relationship between queer and club music and how it's been somewhat co-opted.  I'll take that over "I realise you like songs with strings in the breakdown and a bass line that has more than three different notes in it"

[SuperBacon]

7 minutes ago, Your Fight Site said:

Also, all these companies laying people off en masse to replace them with A.I. workflows or whatever, are also going to find they’ll have no one to sell their products and services to if a large percentage of the population is unemployed because “A.I. took der jerbs”.

Maybe B2B sales just becomes Bot2Bot?

[Sphinx]

Definitely, my appreciation for art, or anything, is better the more I know about it. I could go in an art gallery and nothing hit because I'm looking at art at a surface level. It could be beautiful in its colour palette, its brush strokes, what medium the art's in etc. But for me, the lightbulb can come on when someone either explains why they like it, or when I find out the backstory of the art and the artist and then I can make connections and empathise and appreciate their story and notice things from that story that might have informed the art.

But I think that, in the future, AI could create a song I could hear in a club and me go 'that's an absolute banger'. I wouldn't know it's AI, much like I might not know who the human artist is of a track that's played in a club.

It's not there yet. But we're what, 2-3 years into ChatGPT being used by the general public? The internet's 30 years or so into being a home consumer product, smartphone's about 20. I don't think it's out of the realms of possibility it becomes more and more refined.

Edited April 24, 2025 by Sphinx

[hallicks]

The only reason generative AI exists is because companies want a cheaper way of doing or creating a thing they would normally have to pay someone for. It doesn’t matter if it might be good, or if it may one day be good. AI generated art being sold commercially means real artists aren’t being paid, which means less real art is getting created, and eventually, fewer and fewer artists. And it shouldn’t need explaining why this is a bad thing. 

[organizedkaos]

The thing that worries me is our expectations of the read world change in line with technology. Statements need to be under 260 characters, the selfie angles for photos, makeup that recreates instagram filters.  Even with music... MIDI was created as a quick means to signal a computer from a electronic piano keyboard. But now a lot of music is locked into MIDI for music generation as that's how a lot of production software works and we're restricted to the western scales of tone/semi tone because the interface only allows that. People don't notice what they're not getting because they expect what they get (yes, this has changed a bit in the last decade)

Also the people selling us the tech really want those changed expectations so they can continue to peddle a lesser product at the same price as before.  Do we actually start to find AI becomes good enough to make a good song or do we just get used to average AI generated songs and our expectations change so it's now "good".  "Good AI customer service" is still less good than a competent human paid a decent wage.

[organizedkaos]

Thinking about it further Generative AI is in itself potentially an example of locking in tech and resetting explanations. I think there's some incredible stuff in how it works and one day it may be seen as a vital leap in creating AI - but I also believe it would work best as the "language centre" of an AI. We know our brains have different systems and areas for different productions. I think a more interesting AI would feed it's "thoughts" into a ChatGPT esque layer and that's how it expresses itself.

The fear is that because people seem to have gotten really into the Generative AI bit is that funding gets centred on that then we stop developing other models and the current AI becomes the base from which further developments grow.  Thus the "language centre" becomes the whole brain and we're forever stuck with hallucinations limitations to the software because we locked ourselves in to one model. It's like sleep talking, it's kinda cool and weird that the words happen and seem to make sense but i'll fucked if I want to only converse with someone in their sleep

[LaGoosh]

1 hour ago, Sphinx said:

But if AI were to 'strategise' how best to create a song within certain parameters, I think it's possible that one day it could generate something that would both surprise and - put next to a quality human creation - could be considered high quality or as emotionally compelling, if not more, as the human version by a large number of people.

You can't strategise good art because it's an expression of humanity. Sometimes there's no clear reason why a piece of art works - it just does. It's chaotic, original and unpredictable which is something AI can never be because it can only 'create' using things that have already been done. Everything it does will always be shallow and empty and disposable by its very nature. And that's not because there's no history/backstory to the artist or whatever you're going on about.

I find your entire mindset around this absolutely depressing beyond words. 

[SuperBacon]

3 hours ago, Sphinx said:

But if AI were to 'strategise' how best to create a song within certain parameters,

There's no logic to music, it's art.

[organizedkaos]

I think there's some interesting stuff to be understood by finding patterns. I did my dissertation in machine learning to detect genre in music and it was interesting to see which factors in the music (like key, or instrumentation) were more likely to determine genre (although my findings were minimal, it was 20 years ago and I half arsed university).

Other stuff like how the four chords of awesome pop up so often in songs it shows how certain relative keys (minor, sub dominant, dominat) are more pleasing to the human ear. 

But I don't want everyone to start writing a song with those chords workng from there. research and logic can lead to better understanding, i'm not convinced they lead to better creation...