• regbin_@lemmy.world
    link
    fedilink
    English
    arrow-up
    5
    ·
    edit-2
    1 year ago

    If you knew the exact terms to get the the AI to recreate something in its training data, it could, 1:1.

    That’s because you told it to. Don’t make it recreate existing art then.

    And if you ask it to create you something new, no matter what parameters you use it will look like a mess of garbage data.

    This is not always true. You can train it on a certain style and a photo of a random object, then have it generate an image of the random object in that style. It will “understand” the concept of a style and an object.

    ultimately all the capitalists running these tools see it as is another method to bring the public under their exclusive and totalitarian control.

    Exactly why I’m not supporting the closed source paid services (Midjourney, ChatGPT, Bing Chat, DALL-E etc.) and instead advocate for open source projects like Stable Diffusion and LLaMA.

    • Omega_Haxors@lemmy.ml
      link
      fedilink
      English
      arrow-up
      2
      arrow-down
      5
      ·
      edit-2
      1 year ago

      That’s because you told it to. Don’t make it recreate existing art then.

      If you took a random concept and explained it to a person they could using their existing knowledge set, draw it somewhat competently. That is because people are able to apply knowledge to make something new. If you told someone to recreate something that already exists, even if they’re a professional, would never be able to recreate it no matter how much time and effort the put into it. AI can do the latter because it’s basically copying, and it can’t do the former because there’s nothing to copy from.

      • regbin_@lemmy.world
        link
        fedilink
        English
        arrow-up
        4
        ·
        1 year ago

        If you took a random concept and explained it to a person they could using their existing knowledge set, draw it somewhat competently. That is because people are able to apply knowledge to make something new.

        Theoretically it can, but it would involve meticulous and proper labeling of each training data. Currently most of the trained data are automatically labeled and they’re not descriptive/verbose enough. I believe the improvements from the latest version of DALL-E is due to OpenAI’s use of a more advanced image labeler.