She's wrong because producing a bunch of floating point numbers in a big file is not copyright infringement
By that logic you could just encrypt anything you want and claim it as your own work. It's just a bunch of numbers but that wouldn't help you in court. Really anything you put in digital form is just a big bunch of numbers, what matters is the video/audio/text that your program produces from those numbers.
She might be right if they were storing those works as part of the AI's data
If it can produce the full text then they are, just in a complicated way.
If they rule in her favor it's going to set a bunch of retarded precedent.
Yeah, the retarded precedent that you can't just steal people's work, pass it off as your own and monetize it just because you used a new algorithm. Which will also save millions of jobs but that's retarded because giant corporation has flashy gadget.
TL;DR: If you don't want to get sued over the output of your chatbot, don't train it on stolen work and monetize it on the public internet.
By that logic you could just encrypt anything you want and claim it as your own work
Only if you released the key to decrypt it. If not, it's just random numbers. But that's not a good comparison because the work is directly recoverable from an encrypted file. Which is not the case for the matrix weights I'm talking about. It could never reproduce, 100%, any of the copyrighted works verbatim. Unless it was overtrained on that particular text, which I am sure it's not.
If it can produce the full text then they are, just in a complicated way.
It can't, though. That's the thing. It absolutely, 100%, cannot in any way reproduce the copyrighted text. If Silverkike is claiming this, she's wrong and will lose.
Yeah, the retarded precedent that you can't just steal people's work
Given it cant' even reproduce the text, it isn't stealing. But again, don't put your stuff on the public internet if you don't want it scraped.
(post is archived)