pouët.net

absolutely yes.

here's a (just for fun) video I recorded the other day while trying out Synplant 2 (it's a relatively simple synth that lets you load a sample, then approximates it via "AI") (cheers to @Twilight for the recommendation).

while you need crowd-sourcing for "natural" inputs, building a network for *procedural* sources can be fully automated.

sure, it's still computationally expensive (50k+ iterations needed for learning) but it's doable even if your name ain't Google or Meta.

p.s.: I have to add that I'm still in the "meh, why let the machine spend countless hours on sth I have to post-edit and can do myself in less time" camp so I put this (todo) on the back burner.

It is "teh future", though. Kudos, Sonic Charge, for leading the pack ! (it's a 10+ year dev effort, though, nothing quite so "easy")

When I first saw Synplant 2 it made me sad. You can either spend years (or your whole life) learning, perfecting your sound design or you can just let Synplant analyze Eno’s sounds and recreate them. Not only you don’t need to understand how these sounds are made, but Synplant doesn’t understand either. Sure, it may sound great to some. Progress. What bothers me about this whole tech is the input / training data. Input / training data is never a number, never an idea, a formula, algorithm. It’s never driven by users ingenuity. Input is always a finished, produced work of a human or a whole crew, while training data is a whole “cultural history” in a sense. Be it images, videos, sounds. All of this data is taken without consent. When using these things you’re not standing on the shoulders of giants, you’re pissing all over the giants. But, let’s put the ethics aside for a sec. Who will be filling the new book of “cultural history” if all creation moves to analyzing what’s there and recreating it. Many of the now iconic sounds came to be by happy accidents. For happy accidents you have to fiddle. Not with parameters of an AI “synth” but with parameters of sound design, with oscillators, routing, fx, feedbacks, whatnot. The first crop of regurgitation came in the way of sampling. But you have to clear the samples or at least credit them. Well, not anymore. Excuse me for not seeing this as progress. Same thing with images.

You know, that's a really big IF.

@Krill:
That’s right! That’s why I (and a lot of others) squirm and fight. To make the “IF” bigger. If we just stand and watch (or even applaud) then there’s no “IF”, it’s dead certain. My personal (of course highly inacurate) prediction is that the internet as you and me know it is going to consist of something like 90% of “AI” created content by this time next year. Or the next. Consequences? I don’t know. Look how it changed Google search by now. Shitty GPT made fake “websites” pop in the top 10. On any topic. It got very hard to find the real good stuff. And this is just the beginning of the real beginning.

Do what you must, even if it's making a really big IF even bigger. :)

One can always hope. In reality however, all chances are our collective squirming makes no difference.

more AI crap of mine (follow me) : https://www.instagram.com/latentspacecadet/

AI is an S curve not exponential.

That's why Tesla self-driving cars are still not Full Self-Driving after all these years.
That's why GPT-4 is not orders of magnitude better than the first chat GPT.
That's why that latest image generation models are not orders of magnitude better than the early stable diffusion 1.4.

Has Generative AI Already Peaked? - Computerphile

4gentE: When I first saw Synplant 2 I though it would be great if it actually worked with any content, since big part of sfx work is not that creative and same goes for any audio and art asset which just needs variations. There is creativity in making that first variation of sound, first grass texture, first rock 3D model and so on, but very few people are going to miss making yet another variation 5.

@sauli:
Sounds like a nice “asset machine” to ease the boring work. I wasn’t thinking of this “genetic variation” it does on a sound. I was thinking more of the “here’s completely produced bass sound sample from a Prince album, now recreate it for me as a patch”. Or: “here’s that iconic Phil’s drum sound that came to existance thanks to a happy accident while recording a P. Gabriel album, now recreate it for me as an editable patch”. That kind of stuff.

PauloFalcao: exactly, for better or for worse. Also performance scales logarithmically to the number of parameters/training data size.

So all this FUD (fear, uncertainty, doubt) people are spreading is kind of like being afraid that at some point super intelligent aliens will arrive and make us our pets. There is definitely non-zero probability it will happen.

Real problem was never that this shit is actually so good that it’s gonna replace us all, problem was always that it’s pure crap. Which will not stop the techbros from shoving it down our throats. OK, now I’m aware that some throats are more than willing to receive this demon spunk. But we ALL are gonna be force-fed.

Lol but why u so dramatic "demon spunk". If you hate tech bros you should be actually happy they invested so much in tech that will be essentially disappointing. It's already clear that promotional Sora videos had significant VFX touch, etc...it's funny how they try to fake it just to get more VC money.

Yeah, I think it's becoming apparent that the technology is hitting a plateau and was way oversold both in terms of future capabilities and efficiency. I recently saw this post and it paints a fun picture (pun intended) as to where things are headed.

I saw that post. Tho withbthevamount of push from management on the rise, I’m affraid I can already imagine this: “Well if an art director cannot use the AI prompters, we’ll just have to get rid of him. He’s getting in the way of progress.” You know, the kind of progress that gave us diesel engines in small city cars.

Disappointing to whom? I see many people still being enchanted by this crap. Beating the drum. Just scroll through this thread, and some other threads too. In fact, please correct me if I’m mistaken @tomkh, but weren’t you one of the proponents/evangelists some half a year ago in another thread?

Like, when they made the first Big Brother (as an example of pure crap) was there widespread disappointment?

People were enchanted by Michael Jackson and 3D movies one time as well.

I actually got heartened by the SORA videos, 'cos that's where they overplayed their hand, and showed that they've hit a limit. There'll still be damage, but now we're mid-ripple where they're creating fake journalists and the exposes are coming out, and everyone is weary, including large companies who want to see the results of the major lawsuits to see if it's even legal to use any of this.

So this dumb bubble could burst in techbro’s faces? Yippee-Ki-Yay! You made my day!

@Gargaj: And then there will be artists who use AI prompts as an starting-image. And they will not fall into the pithole like the prompters above.

@gaspode
Ah yes, the quislings. ;-)

I do like ML/AI/AGI, yes. But I don't remember ever drolling over those toy products. It's not my fault that some people have tendencies to be greedy and delusional.



What do you mean "were"? But you are right, recently every big-ego director has switched to IMAX 70mm non-sense. IMHO proper 3d is way better!

Offtopic, I know. But I just gotta ask: you’re joking, right?

Not joking. I very much prefer crisp digital HFR 3d. 3d got bad rep.due to lots of horrible lazy movies featuring it, but if done right it can be really immersive.