• 0 Posts
  • 91 Comments
Joined 1 year ago
cake
Cake day: June 4th, 2023

help-circle







  • Yes, but how does that negate its usefulness as a tool or a foundation to start from? I never made any assertion that AI is able to make connections or possess any sort of creativity.

    It is useful. Never said it wasn’t. I’m pointing out problematic uses of an otherwise good tool.

    Maybe it’s easier to think about this through the lens of the end goal. We want good art to exist. We want good art to continue being produced for the foreseeable future. What inhibits this from happening? If artists stop producing art and AI can’t replace them, then we stop getting art. The point about current AI not being able to create the kind of art we care about is that we still need human artists. So how do we ensure that human artists continue producing? By making sure they get properly compensated for value they produce and that their work does not get used in a way that they don’t like. I’m personally not a fan of forcing people to work, so my preferred solution would be to give artists what they want in exchange for their work.

    There’s a common saying that there is no such thing as an original story, because all fiction builds on other fiction. Can you see how that would apply here? Just because thing A and thing B exist doesn’t mean that thing C cannot possibly be interesting or substantially different. The brainstorming potential of an AI with a significant dataset seems functionally identical to an artist searching for references on Google (or Pixiv).

    I’m not sure if I understand this correctly. Are you saying that an interpolation between two existing artworks can still make interesting artwork? If so, then yes, but if that’s all you’re doing, it severely limits the space of art that you have access to compared to something that also interpolates with a human being’s unique life experiences and is also capable of extrapolating by optimizing for the emotional cost function.


  • Whatever you decide to call it, the problem exists.

    When you trace or use existing art as reference, you’re using this to learn and not passing it off as your own design. Equivalently, when training an AI model, it’s the same. I don’t think the training part is a problem. The problem comes when producing work. A generative model will only produce things that are essentially interpolations of artworks it has trained on. A human artist interpolates between artworks they have seen from other artists, as well as their own lived experiences, and extrapolate by evaluating how some more avant garde elements tickle their emotions. Herein lies the argument that generative AI in its current state doesn’t produce anything novel and just regurgitates what it has seen.

    There’s also the problem of “putting words in someone else’s mouth”. Everyone has a unique art style (to a certain extent), just like how everyone has a unique writing style, or a unique voice. I’ll speak on voice first since more of us can relate to that. Having someone copy your voice to make it say things you did not say is something many will be very uncomfortable with. To an artist, art style and writing styles are the same.

    The economic side is also a problem. And while I don’t expect generative AI to go away, it can be done in a way that is fair to the people whose work have made it possible and allows them to continue doing what they do. We should be striving towards that.




  • AI in the public space is a joke. It is all based off of the transformers library in one form or another. Go read the introduction page for the Transformers documentation on hugging face. It clearly states that it is incomplete and its intended use is as a simplified example code only. AI is enormously complex in its real capabilities. Most of the issues are due to the simplifications made to allow the ignorant public to use it.

    Which page/passage are you referring to? I’m pretty sure you’re misreading or misinterpreting something because Huggingface has a good chunk of the state of the art models implemented. They’re complex in capabilities, but the implementations are incredibly simple, and that’s part of why it’s taken off the way it has.