5mo ago

Court documents show Meta torrented terabytes of pirated books to train AI models, and employees wouldn't stop emailing each other about it: 'Torrenting from a corporate laptop doesn't feel right'

www.pcgamer.com

Court documents show not only did Meta torrent terabytes of pirated books to train AI models, employees wouldn't stop emailing each other about it: 'Torrenting from a corporate laptop doesn't feel right'

17 comments

And yet again from Corporate America: "Rules for thee but not for me!"
- Corporate America maybe. Meta is an international company and there are also a lot of international publishing companies.
  I could definitely see this leading to a lawsuit in other parts of the world. This is unquestionably a violation of the Berne Convention.
  https://www.wipo.int/treaties/en/ip/berne/
- 81.7 terabytes of books
  ('more than', but we'll use this figure)
  ➗ divided by ➗
  1 mebibyte per work
  (the samples of digital novels i used here to estimate size per work were around 500kb each; i took that and doubled it)
  ✖️ multiplied by ✖️
  $250,000 maximum fine (per infringement)
  🟰 equals 🟰
  $19.48 trillion maximum fine.
  
  I'd be fine with meta having to pay that 👍
  
  A fine like that will last about as long as it takes for ~~Donvict~~ FElon to sign an order making training AI fall under "fair use," unfortunately.
  Or, just as likely, the individuals will be the ones found responsible rather than the corporation.
- If they get sued and it gets to the discovery phase, they might have to actually provide all training data used, and I'm almost certain there is a lot more stolen content in there. But sadly, it will probably not come to that, because they have money.
- It's a win-win either way.
  Although I'm almost rooting for Meta just so that copyright is weakened at all. The damage copyright has done to culture and science might even exceed Meta's damage to society.
Any of those books published by Nintendo, per chance?
- Nintendo is going to have to send out more than just Luigi. There's a lot of plumbing to be done!
Can I do this on my own computer? I’m training an AI I swear.
- Believe it or not, straight to jail
- Me too but with movies and TV shows
  
  Me three, but with, uh .... Porn
I really don't understand how LLM models aren't considered derivative works of the material they were created from under copyright law.
I would love to see Meta sued by anyone for this. They'd win, but that could theoretically could be used as precedence for individuals.
I love that they used a stock photo of someone seeing goatse for the first time as the article pic 😄

17 comments