←back to thread

1503 points participant3 | 1 comments | | HN request time: 0s | source
Show context
CamperBob2 ◴[] No.43574347[source]
And? What's the model supposed to do? It's just doing what many human artists would do, if they're not explicitly being paid to create new IP.

If infringement is happening, it arguably doesn't happen when an infringing work product is generated (or regurgitated, or whatever you want to call it.) Much less when the model is trained. It's when the output is used commercially -- by a human -- that the liability should rightfully attach.

And it should attach to the human, not the tool.

replies(9): >>43574359 #>>43574421 #>>43574494 #>>43574501 #>>43574545 #>>43574555 #>>43574573 #>>43574601 #>>43575140 #
4ndrewl ◴[] No.43574421[source]
Assuming you can identify it's someone else's IP. Clearly these are hugely contrived examples, but what about text or code that you might not be as familiar with?
replies(2): >>43574452 #>>43575756 #
1. alabastervlog ◴[] No.43575756[source]
https://spiderrobinson.com/melancholyelephants.html

Given enough time (... a surprisingly short amount) and enough people creating art (say, about as many as we have had for the last couple hundred years) and indefinitely-long-lived recording, plus very-long copyright terms, the inevitable result is that it's functionally impossible to create anything within the space of "things people like" that's not violating copyright, for any but the strictest definitions of what constitutes copying.

The short story treats of music, but it's easy to see how visual arts and fiction-writing and the rest get at least extremely crowded in short order under those circumstances.