• @[email protected]
    link
    fedilink
    English
    312 months ago

    I’d love for this to be true, but I’d need to see some proof. It feels like wish-fulfillment.

    • @SaucySnake
      link
      492 months ago

      https://arxiv.org/abs/2306.07899 here’s a paper that found that one of the biggest sources for LLM training data is corrupted by people using AI to complete the tasks. There are plenty of papers out there that show the effects of this, which they call “model collapse”.

    • @FMT99
      link
      152 months ago

      Same. I keep hearing folks mention this but it’s not like AI developers aren’t aware of this (apart from a bunch of shitty startups that would fail no matter what) One way to deal with it for example is Microsoft is shelling out so much for “pre-AI” datasets (Reddit) but I’m sure there’s a lot more of those kinds of initiatives.

      Google on the other hand is going to be hard pressed to deal with the ever increasing deluge of AI spam.

      • @Ultraviolet
        link
        English
        262 months ago

        That’s a way to deal with it, but in the long term, “pre-AI” becomes a longer and longer time ago, and less and less useful for any practical purposes.

      • I Cast Fist
        link
        fedilink
        English
        12 months ago

        Google on the other hand is going to be hard pressed to deal with the ever increasing deluge of AI spam.

        Given how they’re one of the main culprits of showing AI spam on the first page, I don’t think they care at all

    • @[email protected]OP
      link
      fedilink
      52 months ago

      While I don’t have definitive proof I’ve seen both AI fanarts that are so good it’s hard to tell if it’s AI or not and AI abominations that are so bad and artificial it want to make you puke so I guess it really depends on the model.

      Funny enough adobe now offers AI generated stock photos that are closer to those abominations rather than anything good. Though if you think about it AI stock art is so pointless… There are already so many stock photos you can choose from. Why would you go out of your way to choose a photo that looks almost the same as regular stock photos but people have 4 arms in it…

      • I Cast Fist
        link
        fedilink
        English
        22 months ago

        Why would you go out of your way to choose a photo that looks almost the same as regular stock photos but people have 4 arms in it…

        You never know what smut some people are writing. Maybe that 4 armed horror was exactly what they needed for their cover.

    • @[email protected]
      link
      fedilink
      English
      32 months ago

      I’ve heard training on synthetic data is fine now. Most datasets are augmented with synthetic data.