• A_Very_Big_Fan@lemmy.world
      link
      fedilink
      English
      arrow-up
      1
      arrow-down
      5
      ·
      edit-2
      7 months ago

      but it is still capable - by your own admission - of doing it

      And if you are comparing LLMs and hammers, you’re just proving how you fundamentally misunderstand what LLMs are and how they work

      And a regular hammer is capable of being used for murder. Which makes calling a hammer that evaporates before it can be used for murder “unethical” ridiculous. You’re deliberately missing the point.

      And it still profits from the unlicensed use of copyrighted works by using such material for its training data

      I just don’t buy this reasoning. If I look at paintings of the Eiffel Tower and then sell my own painting of the building, I’m not violating the copyright of any of the original painters unless what I paint is so similar to one of theirs that it violates fair use.

      it is a composite of copyrighted work

      It’s stable diffusion, not a composite. But even if they were composites, I’m allowed to shred a magazine and make a composite image of something else. It’s fair use until I use those pieces to create a copyrighted image.

            • A_Very_Big_Fan@lemmy.world
              link
              fedilink
              English
              arrow-up
              1
              arrow-down
              4
              ·
              edit-2
              6 months ago

              AI, unlike a human, cannot create unique works of art. it can old produce an algorithmically-derived malange of its source-data recomposited in novel forms

              Find me a single sentence in that entire article that suggests AI art is composites of source data

              You can’t, because how it actually works is wildly different than how you want to believe it works.

                • A_Very_Big_Fan@lemmy.world
                  link
                  fedilink
                  English
                  arrow-up
                  1
                  arrow-down
                  3
                  ·
                  edit-2
                  6 months ago

                  Mhm, I’m sure that’s why you couldn’t find a single sentence about compositing images

                  DALL·E 2 uses a diffusion model conditioned on CLIP image embeddings, which, during inference, are generated from CLIP text embeddings by a prior model.

                  You’re either projecting or being dishonest