Whoops! Facebook trained Llama AI model on pirate site LibGen, with Zuckerberg’s OK

Facebook is claiming its use of copyrighted works for training was fair use.

Yeah, teens downloading Metallica albums off Napster twenty years ago to listen in their parent’s basement was a serious crime, but Big Corp Inc. feeding books to the copyright erasing machine is somehow fair use. The gall.

here’s some interesting context on the class action:

They wanted an expert who would state that 3D models aren't worth anything because they are so easy to make. Evidently Shmeta and an ivy league school we will call "Schmarvard" had scraped data illegally from a certain company's online library and used it to to train their AI...

this fucking bizarro “your work is worthless so no we won’t stop using it” routine is something I keep seeing from both the companies involved in generative AI and their defenders. earlier on it was the claim that human creativity didn’t exist or was exhausted sometime in the glorious past, which got Altman & Co called fascists and made it hard for them to pretend they don’t hate artists. now the idea is that somehow the existence of easy creative work means that creative work in general (whether easy or hard) has no value and can be freely stolen (by corporations only, it’s still a crime when we do it).

not that we need it around here, but consider this a reminder to never use generative AI anywhere in your creative workflow. not only is it trained using stolen work, but making a generative AI element part of your work proves to these companies that your work was created “easily” (in spite of all proof to the contrary) and itself deserves to be stolen.