AI is stealing writers’ words and jobs…

[Scribe]

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

www.technologyreview.com

I'm sure it will be fine. /s

 

[overgeeked]

Not even trying to hide it.

Artists fume as new Midjourney v6 feature suggests their names for prompts

Midjourney v6 gets a new turbo mode and an improved image description function. The latter frustrates artists because it also suggests artist names for your prompt.

the-decoder.com

 

[trappedslider]

Five of this year’s Pulitzer finalists are AI-powered

Two of journalism's most prestigious prizes — the Pulitzers and the Polk awards — on how they're thinking about entrants using generative AI.

www.niemanlab.org

 

[trappedslider]

Nvidia sued over AI training data as copyright clashes continue

Copyright suits over AI training data reportedly decreasing AI transparency.

arstechnica.com

 

[Jfdlsjfd]

Emad Mostaque, CEO of Stability AI, said on twitter that the dataset for SD3 was fully respecting opt-out "unlike anyone else", after mentionning their training method on arstechnica. Since the other major players in the field (Dall-E and MJ) don't communicate on this, maybe he's privy to insider info, maybe he's just commenting on their lack of transparency. If confirmed, it's good to have another major player (along with Adobe) who has an EU-compliant generative system.

 

[trappedslider]

I found this and have sent it to a couple of folks via private but I thought everyone should take a read. It was most recently updated on the 14th, I was not the original creator of the document. I will further note it does not as far as i have read related to chatgpt only to to AI art.

AI Art and Copyright, Steelman & Responses

docs.google.com

 

[Ryujin]

Emad Mostaque, CEO of Stability AI, said on twitter that the dataset for SD3 was fully respecting opt-out "unlike anyone else", after mentionning their training method on arstechnica. Since the other major players in the field (Dall-E and MJ) don't communicate on this, maybe he's privy to insider info, maybe he's just commenting on their lack of transparency. If confirmed, it's good to have another major player (along with Adobe) who has an EU-compliant generative system.

It should be an opt-in model, not opt-out. Not knowing that you have to opt-out of something means that you don't do it. On the other hand the robots.txt of my server is...

User-agent: *
Disallow: /

..., which is a deny all, but I'm constantly having crawls on my site. That's an unambiguous, global opt-out, but it doesn't stop those who don't care about being ethical.

 

[Jfdlsjfd]

It should be an opt-in model, not opt-out. Not knowing that you have to opt-out of something means that you don't do it.

On the other hand, there is no reason to think that copyright extend the monopoly of the author to training AI models. So someone unaware wouldn't assume it is protected by non-existent laws, or laws he doesn't know. Copyright was a protection against verbatim reproduction of (parts) of intellectual property, without mention of AI training, mostly because AI training didn't exist. The public benefit of having it go all the way in favour of monopoly (forbidding the use of copyrighted material) or the other extreme of allowing it without restriction were deemed inferior to the public benefit derived from the happy middle ground of allowing it without restriction for academic or cultural purpose and with opt-out for commercial use.

 

[trappedslider]

Stable Diffusion just hit the 330 million downloads mark.

 

[Ryujin]

On the other hand, there is no reason to think that copyright extend the monopoly of the author to training AI models. So someone unaware wouldn't assume it is protected by non-existent laws, or laws he doesn't know. Copyright was a protection against verbatim reproduction of (parts) of intellectual property, without mention of AI training, mostly because AI training didn't exist. The public benefit of having it go all the way in favour of monopoly (forbidding the use of copyrighted material) or the other extreme of allowing it without restriction were deemed inferior to the public benefit derived from the happy middle ground of allowing it without restriction for academic or cultural purpose and with opt-out for commercial use.

Well that's the discussion here (and elsewhere), isn't it? AI can and does reproduce works that have been read into it. Is it a case of "infinite monkeys", or is it because the works exist in their complete form in the generative AI? Does copyright apply or doesn't it? If it reproduces the works that were used to train it then they could be assumed to exist within it. To me, that breaches copyright. Let's see where the law comes down on it and the "public benefit" is debatable. Damage to the individual has to be weighed against it, even if benefit exists.