Email or username:

Password:

Forgot your password?
All posts David's posts Post Back to profile
Top-level
David Revoy

@sepia These questions are in loop in the history of modern art and contemporary art, and it is a fascinating philosophical adventure to read about it and forge a philosophy. More than 100 years of artists tried to answer that.

About getting used to AI, I embraced it in September ( read framapiaf.org/@davidrevoy/1089 ) but that was before all artists − thanks to website like haveibeentrained.com/ − saw their artworks being used without their consent.

*That* changed everything.
14 Dec 2022 at 23:50 | Open on framapiaf.org
16 comments | Expand all CWs
Yiming Wu ✅ Use OurPaint

@davidrevoy @sepia everyone's data eventually will get scraped so I'm not worried about that. I don't even bother to check if mines were being scrapped.

Just do the thing you do, ai doesnt have the personality and the life experiences to make the same thing you make

14 Dec 2022 at 23:56 | Open on mastodon.art
Tanner Spaw

@chengdulittlea @davidrevoy @sepia agreed on both points, even as someone that is messing with ai art. But I would still want to see a model that's just public domain stuff and the rest being opt-in.

15 Dec 2022 at 0:03 | Open on social.vivaldi.net
David Revoy

@tannerspaw Same, I wish a model existed trained exclusively on CC-0 and Public Domain resources.

@chengdulittlea @sepia

15 Dec 2022 at 0:42 | Open on framapiaf.org
David Revoy

@chengdulittlea I know, but it's not the A.I with its little legs who have decided by a beautiful morning to scrape the web for art alone. 😅

No: it's a team of dev − eg. working at LAION laion.ai/faq/ − who decided to scrape all URL and ALT text, enrich the database with their description, curate the result (eg. remove the watermark, keep the bias of stereotype work/gender/race). Political choices here.

I just hate that.

(Edit: I removed a Ctrl+V error😅 )
@sepia

@chengdulittlea I know, but it's not the A.I with its little legs who have decided by a beautiful morning to scrape the web for art alone. 😅

No: it's a team of dev − eg. working at LAION laion.ai/faq/ − who decided to scrape all URL and ALT text, enrich the database with their description, curate the result (eg. remove the watermark, keep the bias of stereotype work/gender/race). Political choices here.
15 Dec 2022 at 0:06 | Open on framapiaf.org
[DATA EXPUNGED]
Tak

@davidrevoy @chengdulittlea @sepia also it claims they didn't scrape anything at all, common crawl is an old project that created dumps of the ~entire internet, they seem to have simply formated whatever can be formatted there into a txt-img paralel corpus.

15 Dec 2022 at 0:48 | Open on scholar.social
Tak

@davidrevoy @chengdulittlea @sepia oh actually the large one does have a detector for watermarked images, and for I believe porn, so that any downstream user could avoid training on such images.

15 Dec 2022 at 0:54 | Open on scholar.social
David Revoy

@takloufer

Ha, right. Please ignore my previous reply, I read this one of your a bit later after sending it. Yes, a Mature+Watermark filtering.

@chengdulittlea @sepia

15 Dec 2022 at 1:53 | Open on framapiaf.org
Ramin Honary

@davidrevoy @chengdulittlea @sepia

So according stable diffusion AIs trained on web-scraped images, ALL doctors are white men wearing blue shirts with neckties, lab coats, and stethoscopes.

AI can solve problems ONLY IF you have good data to train it, and acquiring is very costly. My professional experience working with AI is that there is hardly ever good data. But companies cut costs, use bad data (garbage in, garbage out) and systemic biases can end up being reinforced.

15 Dec 2022 at 2:19 | Open on emacs.ch
Natasha Nox 🇺🇦🇵🇸

@davidrevoy @sepia "Register an account to opt-out your images from Stable Diffusion V3"

This is the most fucking upside-down logic I've ever seen, and definitely 100% illegal basically anywhere on the planet. However this is the internet - it will be used anyway, and with the modern economy also using data in immoral amounts things won't change in the near future. 😔

15 Dec 2022 at 0:01 | Open on chaos.social
Liberonscien :VeryQueer:

@Natanox @davidrevoy @sepia I’m reminded of how Internet Archive operates on an Opt-Out model. Do y’all have a problem with Internet Archive? After all, it scrapes stuff too.

15 Dec 2022 at 1:26 | Open on eightpoint.app
Natasha Nox 🇺🇦🇵🇸

@Liberonscien @davidrevoy @sepia Valid argument. However the Internet Archive does not process or manipulate them in any capacity other than to make sure they stay accessible. Neither do they profit from them nor replicate anything to profit from.

From a legal standpoint it might be comparable. Not from a moral one though.

15 Dec 2022 at 1:30 | Open on chaos.social
David Revoy

@Natanox @Liberonscien

Yes, very interesting point about Internet Archive, and fair reply from Natasha about the purpose, finality and moral.

@sepia

15 Dec 2022 at 1:59 | Open on framapiaf.org
Owl!🦉

@davidrevoy @sepia
I used to draw with stroke-based rendering that i saw from some dude somewhat like 5 or 6 years ago (even can't remember his nickname now), and i never asked his consent. Should i now kill myself for the sake of justice?

15 Dec 2022 at 0:54 | Open on lor.sh
David Revoy

@lonelyowl First, you haven't made any commercial benefit out of it, nor have you pretended to replace the job of the said "dude". 🤣

And also, please don't compare your own process of learning and practicing with your affection for a style or an artist at a cultural time to a software feed with a long list of URL and description batch downloaded and curated by a team of worker (eg. no Mature, no Watermark, and more other choices). It's not intelligent, it's a tool.

@sepia

15 Dec 2022 at 2:11 | Open on framapiaf.org
Owl!🦉

@davidrevoy @sepia
> you haven't made any commercial benefit out of it

Well, in general, I tried to draw for money, but the process of creating a single simple picture was incredibly long, one time i spent four days without sleep and get 15 bux for it. I assume that even african kids in cobalt mines get paid more. Now i draw only for the lulz and because of boredom.

However, other people were much more lucky in this regard. We have enough of clones of popular artists like kuvshin, sakimichan and others who make a lot of money by patreon and comissions, and no one soul saw a problem within it.

> And also, please don't compare your own process of learning

I'm lack of technical knowledge to have a strong opinion about it, but i have an observation how a system that never saw drawings with Taylor Swift successfully adopted real life Taylor's appearance to generic art rendering style.
(I merged checkpoints of novelai and Taylor model from someone from reddit)

Claiming that this relatively primitive system can reason would be too naive, but it seems to be capable of more than just brainless copying.

Anyway, complaining that other people's pictures were used in training without consent looks like absolutely fantastic greed, comparable to that demonstrated by Disney and other media corporations, because even DMCA leaves ways to use other people's work to create derivative works, as long as they are not completely brainless plagiarism and/or publishing without changes. If you ignore this fact, then it turns out that that dude has the right to burn part of my brain so that I forget about his drawings.

The only thing that is really sad about this whole story is that it looks like another profession will die as a result of automation, and there is still nothing like UBI anywhere in the world. In this regard, I am extremely interested to know what is the point of educating economists if we going ignore them anyway.

@davidrevoy @sepia
> you haven't made any commercial benefit out of it

Well, in general, I tried to draw for money, but the process of creating a single simple picture was incredibly long, one time i spent four days without sleep and get 15 bux for it. I assume that even african kids in cobalt mines get paid more. Now i draw only for the lulz and because of boredom.
15 Dec 2022 at 5:42 | Open on lor.sh
Néstor Arellano

@davidrevoy @sepia

Hi, David. You can no opt-out your images in haveibeentrained.com/. This will prevent, at least, Stable Diffusion 3 to be trained with your works.

Cheers. :)

15 Dec 2022 at 15:45 | Open on mastodon.social
Go Up