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

The AI Unbundling

stratechery.com

Post reply on HN
Print view View original

161–170 of 207 posts

Re: The AI Unbundling

#161

Post by sinenomine »

rpdillon wrote: post #131
clankyclanker wrote: post #119

Earlier quoted context omitted.

Sure, but eventually we’re going to hit on environmental and cost-effective power limits of training, and it’s not worth the cost to train the model. AFAIUI, that’s part of the point that Gebru was trying to make before she was fired.

This makes me wonder what the costs of training an actual human artist are. Are AIs less efficient?

You don't have to retrain an AI to spin up another instance - just download a 4gb weight file via a magnet link floating around the internet and run some python code in terminal on your old PC. This kills the comparison.

And training a real living breathing person in a rich OECD country is going to be costly - no offense meant, I'm actually not from OECD.

Re: The AI Unbundling

#162

Post by sinenomine »

notahacker wrote: post #151
sinenomine wrote: post #101

Earlier quoted context omitted.

Every instagram, facebook, and tiktok photo with associated text data is a potential pair for training. In the smartphone age, the case for data hunger looks pretty weak.

But not as weak as the case that the route to production grade commercial art is reached via biasing the training dataset more towards sloppy social media images...

There is no problem here, for any moderately "in" person it's obvious you can bias the model towards aesthetics by concentrating the highly liked images in the dataset. And if it isn't enough, just ask your audience to sometimes rate the aesthetics of the image and use this as a signal for dataset curation.

Artistic styles are often just thin semantic filters over the base 3d geometry that can be learned from photos, and learning these shouldn't require many examples.

Re: The AI Unbundling

#163

Post by nullc »

gizmo wrote: post #33

Although we talk a lot about disruption, only very few technologies are truly disruptive. You can tell by the panic and awe in the air whether you're dealing with real disruption or incremental change. Dropbox made filesharing easier. It's a good product, but not disruptive. Nobody panicked that Dropbox would make their job redundant. Uber was hard on the taxi industry, but fundamentally you still have drivers taking…

> People will insist that SD art isn't real art.

For 99.999% of art that is commercially relevant no one cares if it's "real" or not.

The problem here isn't necessarily that these tools change the story for "real" art, it's that they change the story for _commercially relevant_ art which has always tended towards uninspired and formulaic.

There are a lot of artists that have hobbies in art they consider meaningful while making a living with commercial art that can be software-displaced.

Re: The AI Unbundling

#164

Post by vitaflo »

mejutoco wrote: post #39
gizmo wrote: post #33

Although we talk a lot about disruption, only very few technologies are truly disruptive. You can tell by the panic and awe in the air whether you're dealing with real disruption or incremental change. Dropbox made filesharing easier. It's a good product, but not disruptive. Nobody panicked that Dropbox would make their job redundant. Uber was hard on the taxi industry, but fundamentally you still have drivers taking…

> People will insist that SD art isn't real art. Artists will fight back, and lose. When talking about Stable Diffusion and art there are usually two different aspects of art. I am not going to try to define art, but sometimes we refer to art as illustration, or stock images (what SD puts in danger) and some as broad modern art. I am no trying to say that one is more valuable than the other, but want to qualify these…

>I am not going to try to define art, but sometimes we refer to art as illustration, or stock images (what SD puts in danger) and some as broad modern art.

A note here, the term "modern art" is usually used to define a period of art produced sometime between 1860-1960. When discussing artwork created today (or since about 1950), most would define that as "contemporary art".

Re: The AI Unbundling

#165

Post by Paracompact »

karaterobot wrote: post #107
mejutoco wrote: post #39

Earlier quoted context omitted.

> People will insist that SD art isn't real art. Artists will fight back, and lose. When talking about Stable Diffusion and art there are usually two different aspects of art. I am not going to try to define art, but sometimes we refer to art as illustration, or stock images (what SD puts in danger) and some as broad modern art. I am no trying to say that one is more valuable than the other, but want to qualify these…

This comment is right on the money in making the distinction between illustration/stock images, and fine art, let alone installation or performance art. It's a distinction not made often enough in these conversations. These functions are performed by different people, for different reasons, and they're used in different ways. Sadly, the market for the kind of art that the these models cannot disrupt is relatively sma…

> One other thing I'll tack on here is that I find it fascinating that the kinds of skills required to be "good" at using these image generation models — "prompt engineering" if you like — are largely different than the ones required to create art from scratch.

As someone intimately-amateurly involved with AI artwork over the past six months, I've come to disagree with the term "prompt engineering" as a description of the skillset. It's not just feeding it a good prompt (which is not that much of a puzzle once you get into it, and is close to being a solved problem in my book), but instead a whole iterative pipeline of human-in-the-loop hacks and touchups in order to make up for the tech's existing faults and unwieldiness. Many aspects of this pipeline, interestingly, are already familiar to digital artists.

I think the specifics of this pipeline will continue to change on a weekly basis until we reach a plateau in the technologies. But, novelty lies at the heart of all art: I don't think the pipeline will ever be replaced by one-shot, completely automated and brainless txt2img processes. Even if the difference between "beginner artist" and "expert artist" shrinks to a hair's breadth because of access to AI, it goes back to that xkcd quote, "Our brains have just one scale, and we resize our experiences to fit." We will continue to instinctually desire and respect the efforts of people who go the extra mile to create something just a little more perfect or a little more original.

Re: The AI Unbundling

#166

Post by jaslkjdla »

dan_mctree wrote: post #157
monkeynotes wrote: post #137

Earlier quoted context omitted.

Art isn't about skilfully painting a tree. That's artistry. Art is more than that, it's about feeling, evoking thoughts, and yeah, commenting on society. > It's always trite, reductive, and utterly unoriginal. This is simply untrue, modern art is not "always" anything. And who are you to judge what is going on in the mind of an artist? I don't know what decade you deem to be "modern" art, but there is a huge amount o…

> but there is a huge amount of 20th - 21st century art that is jaw drop What would be some examples of that? Not disagreeing, would just like to have my jaw drop

[dead]

Re: The AI Unbundling

#167

Post by Paracompact »

ben_w wrote: post #136
rco8786 wrote: post #60

Earlier quoted context omitted.

Worth noting that this is also currently on the HN front page: https://waxy.org/2022/09/online-art-communities-begin-bannin...

Huh, was not expecting Fur Affinity to be a poster child for such a ban. The surprise is not because it’s furry — the art scene is such a large fraction of the furry economy that the introduction of AI art is approximately equivalent to introducing modern farming to, IDK, the 1850s or something — but rather because of that site’s lack of reputation with regards to technology. Although, perhaps that explains why the q…

This was actually predictable, if one was listening to the buzz around the different communities when the "this waifu/pony/fursona/etc. does not exist" generators were first introduced. Every community seemed to love the spectacle, except for the furry community, who were noticeably more on edge about it.

I still don't really know what to attribute the attitude to, but it seems to be a pattern.

Re: The AI Unbundling

#168

Post by Paracompact »

simiones wrote: post #73
_aavaa_ wrote: post #61

Earlier quoted context omitted.

> I don't think there's been any industry that's been ended by AI yet, and yet people are strangely confident that art is going to be the first. Technology is making something that used to take a lot of practice and skill be accesible to those without any of it. A monkey can now draw two ovals, label it an owl, and run an image-to-image conversion with Stable Diffusion to get a pretty good sketch of an owl [1]. Is it…

But if I want a good picture of an owl, I Google "owl" and get many more options than I could possibly ever have time to pick from. Stable Diffusion is essentially doing the same thing as Google, except presenting a kind of average result instead of showing me all the results in its DB. Now, this may actually be helpful in that it gets around copyright claims - but that's the only real difference.

Go ahead and get me a photo off Google images of an alpaca in a suit playing chess in vibrant digital painting style.

Without meaning to sounding rude about it... I'll wait.

Re: The AI Unbundling

#169

Post by idontpost »

echelon wrote: post #156
idontpost wrote: post #117

Earlier quoted context omitted.

Why would I want that? Even if I like Taylor Swift, what would be the point?

Today you're a Taylor Swift fan and you want more of her music faster than she can deliver. Maybe you make it yourself and share with friends, and it goes viral. She might even get on stage and sing it. Tomorrow, you want more Taylor Swift and you tell the computer to give it to you. It delivers with no human intervention whatsoever. You love it, but want it a bit more country - so be it. Piano and acoustic. Got it.…

You don't get it.

> A never-ending stream of novel music that tailors itself to you.

That's worthless. A firehose of novelty is almost immediately no longer novel.

Why would anyone want this?

It's the same problem already with streaming services and their own series. They're all already joyless and immediately forgettable. They have less and less value.

Your breathlessly described "amazing transformation" sounds incredibly boring and pointless.

Re: The AI Unbundling

#170

Post by Jevon23 »

Paracompact wrote: post #168
simiones wrote: post #73

Earlier quoted context omitted.

But if I want a good picture of an owl, I Google "owl" and get many more options than I could possibly ever have time to pick from. Stable Diffusion is essentially doing the same thing as Google, except presenting a kind of average result instead of showing me all the results in its DB. Now, this may actually be helpful in that it gets around copyright claims - but that's the only real difference.

Go ahead and get me a photo off Google images of an alpaca in a suit playing chess in vibrant digital painting style. Without meaning to sounding rude about it... I'll wait.

I'd be curious to see if you could get that from the AI as well.

I tried generating that exact prompt a few times at theartbutton.ai and all the results were nonsensical.

For example: https://theartbutton.ai/image/OW1HZLfhjg6DFvJtk4vQZzUYqI7pGG...

Post reply on HN