Hacker News Reader

Comment by tern | Hacker News Reader

The fact that LLMs compute an average of human culture is more apparent in music AIs than any other medium. You cannot get these things to do anything original, same as with images, designs and creative writing—and it's not an "intelligence" problem.

I'm not sure if this is solvable, but I think it should be a bigger research topic. If anyone knows of any papers on this, I haven't found one yet (not sure what to search for).

View on HN

Any musician will tell you most music isn't original. You just don't have the ear for it. You telling me Green Day's songs are all unique and original? Even jazz uses 6-2-5-1's over and over. Unless you only listen to avante garde prog rock or something, all of music is derivative. And that's ok. Every song being a unique snowflake isn't important. If you like it or it makes you feel something, that's all that matters.

All books are derivative too, the English ones use the same 26 characters over and over.

You're intentionally missing the point, but ok.

The 26 characters with a thousand @font-faces.

It's actually surprising how many ways and styles humans invented to draw the same letters; that amount of talent and creativity AI models do not have, at least now.

> If you like it or it makes you feel something, that's all that matters.

You’re right!

> Even jazz uses 6-2-5-1's over and over.

You’re not even wrong! I wonder if jazz does anything else besides that?

Green Day (was) definitely original. The level of unoriginality I'm talking about goes way beyond "pop music is the same old chords from the 50s" or "the 70s were the golden age and everything since then is derivative."

What's the basis for this? Unfortunately it's hard to describe, but I've listened to a wide variety of popular and niche genres my whole life with a specific eye toward appreciating all the different ways people appreciate music and I know when something feels new.

Even most (or all?) pop music feels new. If it wasn't, I don't think it would be popular. Sure, it's all derivative, but what makes music enjoyable is when it combines influences in a fresh way.

"French house polka" achieved by doing a statistical blend of the two genres just isn't that interesting—it misses so many ways that things are combined by artists—specific details of the texture of the sound, how it's recorded, cultural references, tons of tiny little influences from specific artists and other genres, etc.

I've tried very specific prompts in Suno and it's not even close to useful for someone who knows what they're doing. The image generators are hardly better—things overwhelmingly trend toward a fixed set of specific styles that are well-represented in the training sets.

This critique falls down in certain areas though. Using tools like Suno to come up with songwriting ideas can be fantastic—as long as you have the taste to curate the outputs and feed it into your own creative process. It's also fantastic for creating vocal samples, and I'm sure it'll crush it for functional music (advertisements, certain types of social media) in short order

If you think song and image generators can't make creative things, the problem is between the seat and keyboard. It is unfortunately that simple. 4o as an image generator can create out of distribution images because it literally works at a conceptual level.

Suno currently is limited architecturally to in distribution components, so trying to create instruments or vocal styles it never heard won't work. The parts that you can work in are a vast and rich creative space.

Tell me you don't know the first thing about making pictures or music without telling me you don't know the first thing about making pictures or music.

:) I'll bite...

For bonus points.. the tail-less crocodile.. https://chatgpt.com/s/m_6815c29c1bf081918c791447441c400e

You just lack imagination, like I said.. there are almost no limits but pixel density and aspect ratio currently.

You're not making art, you're running prompts through a meme picture generator.

Like I said, you know nothing about making art. (Not a big deal, not everybody needs to.)

In an alternative parallel universe we might have gotten actual AI tools for making art - these would be various infilling plugins, basically. ("Make me a Photoshop brush that paints pine-needle looking spiky things", that sort of thing.)

What we got in this reality is not that, though. Meme making of the sort you posted is not art.

(That said, making memes, especially of the spicy and/or topical kind, is the one niche that generative AI is good at and nails perfectly.)

I mean I did have a meme or two in there to show the range. You clearly didn't look at it if you think it is all memes. Maybe think hard about what art really is.. a brush is a brush..

I don't have to justify my art to anyone. The magic is.. you can't choose what is art. You can choose what YOU think is art and that's fine. However you can't dictate to the world what is art.

I agree you don’t have to justify your art. But which was not a meme? I clicked many and found only memes.

Arguing about what "art" is is a tale as old as time, relax. Your own perspective is not the only valid one that can exist.

That's not the AI creating art though, that's you using 4o like a pencil. The ideas / creativity came from you, not the AI.

P.S. Do you mind sharing the prompt you used to do the crocodile?

Sure... I mean it isn't a single prompt.. it is always a process.

https://chatgpt.com/share/681855ce-a9d4-8004-9a3a-deb19994d8...

Edit: I was really just testing to see how well it could do layout and out of distrobution images. This was right when it came out and I was trying to see what the limitations were.

Nah he's right. Gatekeeping art and "the process" is as old as time.

https://youtu.be/JdxkVQy7QLM?si=_lkqbjVZRuPVjOkL

https://youtu.be/_PC6jwoHyOk?si=imjfG9khOI-hsJA0

https://youtu.be/Fkf4QYTfIxA?si=j-Fo8HN59xCbteD2

Etc etc

My favorite: https://www.youtube.com/watch?v=5pidokakU4I

This raises deep questions: What even is creativity? How does taste emerge? Is it just the accumulation of your experiences and your genetics, or is there something else to it? If it's just experiences and genetics, AI will probably outpace humans at some point, because these are just very sophisticated patterns. And if it's not, we would have to assume something metaphysical, for which we do not have evidence.

This is a really fascinating topic, and I think it might give us new insight into the human condition. I'm excited to see where this leads us.

I like this take. if LLMs = average of human culture like tern said, then taste = the ability to create and recognize the above average (and best), right?

can AI recognize what is best? can AI create what is recognized as best?

(you know how vast majority of humans think they are above average drivers?)

I think that begs the question, what is 'average' or 'best'? It's music enjoyed by many people. Why do they enjoy it? Because it matches their taste in music. I think we're ultimately going to get super-personalised recommendations and even original artworks tailored exactly to what we like. The true challenge in implementing this will be to find a way to 'generalise taste adaptation'. Creating a model that can adapt to any person's taste based on as little information as possible. I think that will give us very deep insights into how our preferences are formed as research progresses.

I'm sometimes scared that life will loose its spark when AI is just able to solve any problem better or faster than humans. I think music generation is scary, because music is often created using your intuition, rather than relying on general principles or strict rules alone. Intuition feels almost magical, because it feels like you somehow just know the right thing to do, without having to reason about it. If AI gained this hard-to-grasp 'intuition', that would reveal that we are just biological machines after all, and intuition is simply a sophisticated hidden pattern.

yes

I think the problem is the lack of a verifier to check the attempt at creativity in a tight loop. Creativity could just be randomness (off the existing average) plus selection done many times over a 30 year career.

That is, you try something new (random) and you, the human, are also the verifier to see whether that random new thing was subjectively good, and then you select based on that.

In this understanding of creativity, creating a new style is a gradual process of evolution (mutation and selection) where your own brain is a necessary part of carrying out the selection. "Did the new idea (mutation) I just tried make me feel good or not (selection)?"

That activity of you being the verifier is effortless and instant, but the AI fundamentally can't tap into the verification and so it has no ability to do the selection step after randomness, and so it is impossible for creativity to emerge no matter the architecture (unless you somehow create a proxy for a human verifier which seems insanely hard).

The only solution I can see to this is to try to simulate this process, seems possible but hard.

I think we should stop research into automating creative labor and study something that actually benefits people.

Reply on news.ycombinator.com