Re:Ask HN: DALL-E was trained on watermarked stock images?
by gfodor on Wednesday August 24, 2022 @11:36AM
(#32580795)
The law fundamentally needs to evolve. Latent embeddings of large corpuses of copyrighted works is something we are going to have to wrangle with more directly, it’s not clear to me how we even ought to want it to work in terms of the rights of the copyright holders for data it was trained on.
With the release of spectral diffusion, arguably the genie is out of the bottle now, so there’s probably a ceiling on how much the law can evolve to claw back any retroactively determined rights to copyright holders.
With the release of spectral diffusion, arguably the genie is out of the bottle now, so there’s probably a ceiling on how much the law can evolve to claw back any retroactively determined rights to copyright holders.