The current state of AI

[BONES]

Weird that these AI systems simultaneously have no agency but are also able to communicate emotion.

Does your DAW have agency? Are you not able to get it to express emotion? Does any physical musical instrument have agency? No, they are just inert bits of wood or metal. It's the human using it that coaxes expression/emotion from it. It's the same with AI, only through a different process.

Whenever I try the tech, I'm surprised just how bad it is, how high the failure rate is, and how limited the quality of the result is.

Then you must be really shit at telling it what you want. I'm not very good at it, either, but my bandmate spends hours a day immersed in it and he's a f**king wizard! So the guy who makes these videos -

When you ask a chatbot basic music questions, you get an answer similar to one from a 13 year old Youtuber who has been playing for no more than a year.

Perhaps that's how the AI interprets the source of the question. Maybe you should try asking it less basic questions?

So, more often than not, something between outright wrong and poorly informed.
Data scavenging does not include evaluation of the data, it does not differentiate between competent expert information and clueless amateurs.

Well, it does, just look at how much better search results have got since AI's been helping out.

A good example of the insane hype is the claim that AI can 'cure cancer'.

Is that a good example, though? It's not something I've seen anywhere and the internet is full of that kind of bullshit, it's not limited to AI. How hard did you have to look for it?

Imagine a common complex human emotion, like how you feel about your grandmother who has recently passed away.
How would you talk about it? How would you express it with lyrics/poetry

Co-Pilot gave me a number of different approaches from a very basic prompt. If I'd added more detail, I'm sure it would have done better but this is OK for literally 10 seconds of work -

1. Raw grief, no filter
Your absence hits like a slammed door.
I keep walking into it,
forehead first,
every morning.
I wasn’t ready for the world
without your hands in it.

2. Quiet, aching minimalism
The house is still.
Your teacup waits.
The light falls
where you used to sit.
I stand in the doorway
and try not to break.

3. Anger at the unfairness
I rage at the sky
for taking you so gently
when it could have left you here
one more season.
The universe shrugs,
and I’m left shouting
into its cold, blank face.

4. Soft, nostalgic, almost whispered
You smelled of lavender
and stories.
Now the garden grows wild
without your voice
to tell it where to bloom.

5. Spiritual, but not sentimental
If there’s a place you’ve gone,
I hope it’s warm.
I hope it’s quiet.
I hope they know
they’re holding someone
the world wasn’t ready to lose.

6. Fragmented, modern, emotionally jagged
Your laugh echoes
in the wrong rooms now.
I keep catching it
in the corners of my mind,
like a glitch
in the code of my grief.

7. Stark, blunt, almost brutal
You’re gone.
And the world didn’t stop.
And I hate that.
And I love you.
And both truths
are knives.

8. Lyrical, dreamy, surreal
You slipped into the night
like a lantern going out —
quiet, sudden,
leaving the dark
to remember your glow.

How would you express it with lyrics/poetry (if there are lyrics), with sounds, with harmony, with subtleties in the timing of notes, with melodic phrasing, with rhythmic phrasing, with instrumentation, with improvisation?

Those are things you would direct the AI to create, just as you'd program a sequencer to put it in your DAW - adding a bit more humanisation, or subtle quantising if you played it in. Editing the patches to give it the gravitas you want. AI can do all those things if you have the patience to deal with it.

The goal of talking about an emotion is to make the person you talk to understand how you feel.

But she's dead so that ain't happenning.

The goal of expressing an emotion through music is to tell an abstract story to yourself based on how you feel, and to give the listener a canvas they can use to project their own stories and emotions on.
To trigger their imagination.

And AI is the prefect tool for that, especially if you can't play an instrument and/or have no musical knowledge. That's what it's designed for. It's not designed to let you control the nitty-gritty so much as the overall feel/emotion of the piece. If you specify chords or keys or time signatures, it will take it under advisement but it's not always going to spit out those things exactly. But if you tell it you're sad that your grandmother is dead and you want to express that sadness through a Country ballad or an choral requiem, it might surprise you how well it does (or it might not, I haven't tried anything so gauche with it).

Reverse testing: record something on a good day, on a day when you play well, create something that works. That you like. That has meaning. That is inspired.
Then try to create an AI prompt to reproduce it.

Well, in the time it would take you to do that, you could get the AI to spit out a hundred iterations, one of which is likely to do as good a job. But the thing is, it'll do it all day long, you don't have to wait until you're n the mood.

[Shabdahbriah]

[...] So [is] the guy who makes these videos -

Indeed... That is really impressive.

[BONES]

He was pumping them out at a rate of one every couple of weeks, all with a similar look. If you wanted to do something like this with 3D animation tools, it would take a team of 100 experienced artists months and months. AI is not really ready to take on CGI work for Hollywood blockbusters but, as with music, if you are happy to go with the AI flow, subtly nudging it in the general direction of what you're trying to achieve, it can do truly wondrous things.

I think that's probably where a lot of the naysayers fail - they are trying to get it to do very specific things but music AIs like Suno aren't set up for that, they are more about the "vibe" of the thing. To a certain extent, you have to be prepared to go along for the ride. Just as with any collaboration, you have to be willing and able to make compromises, to let the AI have its head and try to guide it in the direction you want to go.

[Shabdahbriah]

Agreed. ^^^ Well said.

[zerocrossing]

"Artificial Intelligence" is not hype, it's accurate. Artificial sugar is not sugar, it behaves like sugar in some contexts for some people....

This cuts both ways.
You argue that if someone thinks a spreadsheet has intelligence, then it does, because that's how that person defines it. That 'intelligence' is purely subjective....

That doesn't seem to be what he's saying. The intelligence is artificial and behaves like genuine intelligence in certain contexts. Your interpretation of that is weird. He's not saying that if someone misunderstands and projects real intelligence onto it then that makes it so.

I view AIs as fantastic correlation machines - great at identifying patterns in their training data and correlating them with other data, but limited because they don'r really understand what anything is.

I think it can still be very useful within those limits if you give it the right prompts, although I've used it very little myself, just a few Hugging Face spaces and some Duck.ai questions. Some of what I got was useful.

If there's anything that is analogous to human thought from AI, it's dreaming. I noticed this when I first started seeing AI trying to render text on things. Ever try to read in a dream? You can't do it, because that part of your brain is down. It ends up looking like some tiny snippets of words, but also odd "letter like" symbols. A lot like how generative AI renders words. The hallucinations... the general weird glitches. It's often got a very dream like vibe, because I think the part of your brain that does conscious thought is doing reason, and AI does not reason. You can brute force it, a bit, but not really. Not like a human.

[procrastive]

If there's anything that is analogous to human thought from AI, it's dreaming. I noticed this when I first started seeing AI trying to render text on things. Ever try to read in a dream? You can't do it, because that part of your brain is down. It ends up looking like some tiny snippets of words, but also odd "letter like" symbols. A lot like how generative AI renders words. The hallucinations... the general weird glitches. It's often got a very dream like vibe, because I think the part of your brain that does conscious thought is doing reason, and AI does not reason. You can brute force it, a bit, but not really. Not like a human.

Interesting! Reminds me of the left-hemisphere, right-brain hemisphere division of the human brain, with the LLMs being more akin to the left hemisphere and generative AI to the right.

Maybe finding a way to combine the two (LLMs and gen-ai), along with some actual experience of the real world, rather than abstract representations like words and images, could bring AI closer to rational thought - an artificial corpus callosum.

[pdxindy]

He was pumping them out at a rate of one every couple of weeks, all with a similar look. If you wanted to do something like this with 3D animation tools, it would take a team of 100 experienced artists months and months. AI is not really ready to take on CGI work for Hollywood blockbusters but, as with music, if you are happy to go with the AI flow, subtly nudging it in the general direction of what you're trying to achieve, it can do truly wondrous things.

The 'render' quality is remarkable!

The tools to have more control over the output using storyboarding and character sheets are improving. I'm curious if AI will eventually become good enough for movie work, or if it will plateau and always be hit or miss.

[BONES]

I think it's good enough in terms of quality now, and the consistency seems to get better and better all the time. It's more the difficulty in getting it to do exactly what you want it to and film directors are nothing if not megalomaniacal tyrants.

[zerocrossing]

I think that's probably where a lot of the naysayers fail - they are trying to get it to do very specific things but music AIs like Suno aren't set up for that, they are more about the "vibe" of the thing. To a certain extent, you have to be prepared to go along for the ride.

I don't know Suno, but I had a few credits left with OpenArt, so I thought I'd see what it could do if I just went "along for the ride."

Garbage. I'll post it later, but it's hilarious garbage. In one clip, the character is holding a paintbrush (why?) and the text tag <reused prop> hovers above it. In another scene, she is cut in half, as if by a magician. Her lower half is just not there. It's supposed to be great at lip sync, but it's horrible, and many times during instrumental parts, her mouth is twitching like she's having a seizure. I'll post it later. I wrote support and asked for a full refund. What was provided was not what was promised.

[zerocrossing]

If there's anything that is analogous to human thought from AI, it's dreaming. I noticed this when I first started seeing AI trying to render text on things. Ever try to read in a dream? You can't do it, because that part of your brain is down. It ends up looking like some tiny snippets of words, but also odd "letter like" symbols. A lot like how generative AI renders words. The hallucinations... the general weird glitches. It's often got a very dream like vibe, because I think the part of your brain that does conscious thought is doing reason, and AI does not reason. You can brute force it, a bit, but not really. Not like a human.

Interesting! Reminds me of the left-hemisphere, right-brain hemisphere division of the human brain, with the LLMs being more akin to the left hemisphere and generative AI to the right.

Maybe finding a way to combine the two (LLMs and gen-ai), along with some actual experience of the real world, rather than abstract representations like words and images, could bring AI closer to rational thought - an artificial corpus callosum.

No, I think LLMs are much the same, but just using text. No true reasoning ability.

I've not turned it off yet, but Gmail now nags me to change the text of my emails. The suggestions are never "this is grammatically wrong, here's how to fix it," but always, "here's a way to remove all personality from this sentence." It's super annoying.

[procrastive]

I don't know if reasoning is possible without any true understaniding of what things are. AIs are only trained on words and labeled imagery. Biological experience is so much richer. I do think LLMs and chatbots are trying (and failing) to simulate reasoning and are quite different from genAI.

Not that it's possible, but I try to avoid all things Google.

[ghettosynth]

I don't know if reasoning is possible without any true understaniding of what things are. AIs are only trained on words and labeled imagery. Biological experience is so much richer. I do think LLMs and chatbots are trying (and failing) to simulate reasoning and are quite different from genAI.

Not that it's possible, but I try to avoid all things Google.

LLMs are a subset of generative AI. That is, they are generative AI, but so are things that are not LLMs.

[procrastive]

LLMs are a subset of generative AI. That is, they are generative AI, but so are things that are not LLMs.

Yes, but AFAIU, LLMs usually use a different architecture, or is that wrong?

[BONES]

I think what he's saying is that Generative AI is an umbrella term that includes LLMs, i.e. Generative AI doesn't refer to a specific architecture. At least that is how I understand it.

I don't know Suno, but I had a few credits left with OpenArt, so I thought I'd see what it could do if I just went "along for the ride." Garbage. I'll post it later, but it's hilarious garbage.

I was talking about music AI, I've never even looked at any of the video AIs. Who has the time? We fall about laughing at what Tunee spits out all the time. The vocals are often so overwrought, and stressing entirely the wrong thing, that you just have to. Or it will stick something in there that is so out of place, so unexpected, that you can't help but laugh. The one that gets me every time, though, is one of the songs where it gets so worked up that the vocals cease being words. It's like glossolalia or primal screams or something but it's absolutely hilarious.

But when you're not just searching for the negative, there's often something useful to be rescued from the hilarity and we're going to use that song with my vox and some proper lyrics derived from other versions of that particular thread.

[stratology]

New article on Arstechnica:

Deezer says 44% of new music uploads are AI-generated, most streams are fraudulent




Quote:
"In a recent update, the company says AI music is approaching half of all new uploads, and most of the supposed listeners of those streams are AI themselves."