Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AI doppelgänger experiment – Part 1: The training

julienposture.substack.com

Post reply on HN
Print view View original

61–70 of 137 posts

Re: AI doppelgänger experiment – Part 1: The training

#61

Post by ninetyninenine »

nonrandomstring wrote: post #37
ninetyninenine wrote: post #23

Earlier quoted context omitted.

You missed the point. You view it as a war against human rights. Copywrite. I view it as a lost war. The war against human meaning. We're living in a bubble. Where we pretend it all still matters and that AI will always be inferior to everything we do forever and for all eternity. All bubbles will eventually pop.

> I view it as a lost war. The war against human meaning. If "human meaning" goes, all else goes with it; Social media, search, education, ambition, creation, innovation... And along with that goes Google, Microsoft, OpenAI, Apple, Meta, Amazon... There's nothing that exists outside that "bubble" of human meaning, Technology must preserve it, or technology dies. Unless you are confirming what many have already said..…

Human meaning won't go out the door.

It's just we will no longer be able to differentiate it from machine generated content.

Machine generated content will become more superior and more prolific such that it will cheapen human meaning.

You'll find people will begin clinging onto the last vestiges of human created things like real painted art or vintage sculptures like they do vinyl records and non-digital books: An exercise in irrationality in attempt to reverse a war already lost.

Perhaps that is what will save human meaning in the end. Self delusion. We will deliberately ignore superior and better content and tell ourselves that human works are and always will be the best. That's a little of what's happening in this thread already.

Re: AI doppelgänger experiment – Part 1: The training

#62

Post by ninetyninenine »

mistermann wrote: post #52
ninetyninenine wrote: post #29

Earlier quoted context omitted.

>Human creativity turns into a point on that curve. All human creativity are points ON the SAME Curve. It doesn't matter what turns what into what into some point on the curve or finding a different "dimension" If you come up with an algorithm that can traverse that curve you've found the algorithm for human creativity. We are close, deadly close, to the end. Especially given the fact that these AI algorithms literal…

"We're at the tippity top of the mountain, but we're only halfway up".

We're at the foot of the mountain. We only just started.

Re: AI doppelgänger experiment – Part 1: The training

#63

Post by monadINtop »

spywaregorilla wrote: post #44
monadINtop wrote: post #33

Earlier quoted context omitted.

Yep, you're right. But idk why you phrased that like it's a rebuttal to my criticism. This is another failure of our economic system, for incentivising the owners of art production to churn out literal mindless slop to cut costs, instead of safe guarding artists and creators for the sake of their intrinsic value to society and humanity. Since, of course, they must eternally seek to reproduce capital to the detriment…

Well your argument is admonishing tech bros as if they don't get it, but generally they do get it completely in this context. The goal here is not replacing great art. It's replacing art assets. Comes off as naive and distracting to me.

The distinction is philosophical. Tech bros consider "art assests" or consumer oriented products as art. In their eyes its capibility as a tool for profit is a downstream effect.

My point was that since their entire purpose in life revolves around developing and refining technology for the reproduction of capital, it reveals their blindness to any deeper aspect of the human condition than what can be sold for a profit. When people post 3d versions of the Mona Lisa or a shitty remix of Monet's water lillies with breathless praise, it's because in their eyes the stylistic reproduction or reinterpetation of the literal shapes and colors on a flat screen is all there is to art, not merely because it demonstrates utility in replacing already exploited sectors of creative workers.

I don't think we disagree about any of this. I think that I'm just making a tangential point to the article in response to a dumb comment, while you are talking about the subject more directly. Otherwise I fail to see what you think is naive, if only because I've seen so many examples of this rot of understanding around the essence of artistic expression from the mouthpieces of silicone valley and our wider, increasingly corporatized, culture.

Re: AI doppelgänger experiment – Part 1: The training

#64

Post by astrange »

jsheard wrote: post #42
deprecative wrote: post #10

Earlier quoted context omitted.

That last part is the issue. Once the material is released publicly there is zero way to prevent it from being used for training. I like the idea of these things but they're predicated on theft and exploitation.

There are active countermeasures in the form of Glaze/Nightshade, but I don't know how effective those turned out to be in practice.

They don't work, and nothing in the category can ever truly work.

Models and adversarial data are equally powerful - you can find an adversarial example for any current model, but you can also train a model that can handle any existing adversarial data.

Any image is a good example of /something/ - at worst it's only an example of itself and irrelevant for any other class. Don't know if there's a name for this principle but it's kind of like a Church-Turing thesis for data.

Re: AI doppelgänger experiment – Part 1: The training

#65

Post by asabla »

jsheard wrote: post #38
vunderba wrote: post #18

Earlier quoted context omitted.

Corporate Memphis seems to have that very "flat almost clip-art" style to it. Generative art is ideal for this type of stock art use case where 90% is simply good enough for the vast majority of situations. I feel like the gig economy (things like Fiverr) are unfortunately going to get absolutely crushed by generative art systems since most of their customers aren't that particular and are likely fine with anything t…

> I feel like the gig economy (things like Fiverr) Predictably this is already happening, it doesn't take much browsing in categories such as tattoo design to spot the hallmarks of AI generated artwork. To add insult to injury they often pretend they aren't using AI, so they can up-charge for more intricate artwork that would take more effort if they weren't really just adding more synonyms for "detailed" to their Mi…

> it doesn't take much browsing in categories such as tattoo design to spot the hallmarks of AI generated artwork

Which is kind of interesting. My first tattoo has its roots from it. Where I used ChatGPT together with DALL-e to express the concept I wanted on my body, and not a direct translation of it.

Worked out pretty well, got a lot easier to discuss how and what things would work and how to apply them.

Re: AI doppelgänger experiment – Part 1: The training

#66

Post by astrange »

spywaregorilla wrote: post #12
voiceblue wrote: post #4

Related: "Invasive Diffusion: How one unwilling illustrator found herself turned into an AI model" [0]. It's astonishingly simple to train both LoRAs and Embeddings nowadays -- if you combine it with tech like ControlNet, you can more or less have direct control over the illustration as well (see this example I converted to Hollie Megert's style with barely any effort [1] using the model from [0]) -- but FWIW I was p…

> However, I do think there needs to be some kind of protection from having your 'style' commoditized without consent, using unauthorized images of your own art. 1) Hire artist to copy the style for 10 images 2) Train model on those problem solved. If you choose "using your images specifically" to steal your style as the hill you want to die on, then you're going to find it's a very useless line of defense.

Style is largely in the eye of the beholder; SD1.5 was used to make a lot of images that looked like Greg Rutkowski, and people presumably believed they looked like his style, but he wasn't in the training set and it was a coincidence.

Of course this is mostly only possible if you limit yourself to looking at digital art. A JPEG of a gigantic abstract art oil painting is pretty different from the real thing.

Re: AI doppelgänger experiment – Part 1: The training

#67

Post by astrange »

mattkevan wrote: post #41

I’ve been using generative AI in my art for a good few years now and have a few thoughts on this. Yes it’s trivial to train a model on a particular style, I’ve done it many times, including on my own art. It can make things that look pretty good! But it can’t make anything great, not without a lot of luck or creative input from the user. It can make something that looks similar to a style, but it can’t explain what m…

> As I understand it, generative AI is a sort of average machine, in that if you ask it for a picture of a dog, it’ll give you the most average result for the prompt, model and whatever randomness there is in the system.

"Generative AI" isn't a single technology. Text and image generation don't actually work the same way (have the same model architecture), which frankly makes it absurd we invented them both at the same time.

But for text at least this isn't true; LLMs have fractal complexity, which is why they work. ("Average" text would just be the letter 'e'.) It's harder to say how diffusion models even work.

It makes them easier to handle if you train them to act like this though, but try playing with a "base model" and you'll see all kinds of inhuman results.

Re: AI doppelgänger experiment – Part 1: The training

#68

Post by astrange »

GaggiX wrote: post #53
mattkevan wrote: post #47

Earlier quoted context omitted.

As a follow on: One of my friends is a professional illustrator and was concerned about the impact of AI on their work. As a demonstration I trained some models on their styles. They were worried right up until they saw the output. Yes, it looked superficially like their work, but the slightest glance showed it to be absolute crap. Anyone who would consider using the AI version of their work over commissioning the ac…

It's possible that your model wasn't great, for my experience, in the past I uploaded an AI image in the same style of a certain artist to a website that organizes anime images, someone tagged it with the artist's tag because they thought it was made by him, also the image I uploaded got much more "likes" than anything made by the artist in months, and it's not like I'm spamming images, that was the only upload. (I l…

Part of the reason for that is those artists are doing commercial art, which means the "style" is imposed on them by the customer/series character designs/etc, and they're not putting their whole ability into it.

It's like how it was easier to automate office workers because they just followed steps in a process and didn't use creativity.

Re: AI doppelgänger experiment – Part 1: The training

#69

Post by astrange »

AndrewKemendo wrote: post #49
zephyrthenoble wrote: post #9

Earlier quoted context omitted.

You are presenting opinion as objective fact for a subjective medium, which is surprising to hear from an artist. This argument, and this article, are not about the "great masters" but about amateur and professional artists who find their efforts easily reproduced and outproduced by AI. These people are not protected by "This is a REAL Picasso" and some of them could lose their financial security because of it. The a…

My takeaway is that as with many of the artists, they are reacting to the philosophical no mans land that cheap computer reproduction is putting them into...From TFA: >As I’m contemplating this Cronenberg-like transformation of the image, I can’t help to be struck by the triviality of my own work. There’s something confronting in facing a computational doppelgänger, something akin to the uncanny valley. I’m surprised…

> This is my point, and you can disagree, most commercial artists are not producing art, they are designing propaganda for corporations.

Tbf I think the artists most threatened are the ones doing fetish or furry commissions for individual clients paying $100-$300.

Corporations need to be able to make revisions to a style or have artists advise them on where to make changes. AI isn't good at that. (Getting better though.)

Re: AI doppelgänger experiment – Part 1: The training

#70

Post by astrange »

ganzuul wrote: post #11
jsheard wrote: post #8

Earlier quoted context omitted.

> because as soon as there's enough work in the new style it can be stolen. This is where the oft-cited comparison to previous advances like the camera breaks down - AI has an ongoing parasitic relationship with the mediums it is displacing. The camera doesn't need a thousand photorealistic paintings of a new subject to be made before it can capture that subject going forward. AI is an unprecedented situation, as far…

This week I completed a sculpture which I intentionally made to be non-photographable by choosing lighting and materials which showed off high reflectivity and high chroma so that screens would not be able to reproduce it. To the human senses it is something that really steals your attention, perhaps because we have becomes so used to the constraints of our reproduction technology and this is not that. The Prussian b…

I always find it philosophically entertaining when I see a material in real life that can't be photographed with a phone like this. (Of course that includes gold foil, mirrors, blue LED concert lights, the sun, the moon… so, pretty common.)

But try making one that can't be videoed or 3D scanned with Polycam.

Post reply on HN