• Lvxferre [he/him]@mander.xyz
    link
    fedilink
    English
    arrow-up
    9
    ·
    edit-2
    5 days ago

    Bingo - this tree is non-existent outside my homeland, so people barely speak about it in English - and odds are that the model was trained with almost no pictures of it. However one of the names you see for it in English is Paraná pine, so it’s modelling it after images of European pines - because odds are those are plenty in its training set.

        • joshchandra@midwest.social
          link
          fedilink
          English
          arrow-up
          1
          ·
          edit-2
          1 day ago

          What I mean is that if people keep making it produce garbage tied to some keyword or phrase and people publish said garbage, that’ll only strengthen AIs’ neural network between the bad data and that keyword, so AI results for such trees will drift even further away from the truth.

          • KeenFlame@feddit.nu
            link
            fedilink
            English
            arrow-up
            1
            ·
            1 day ago

            Publishing fake data that outweighs the data on the real plant is a way, but that doesn’t require a plant, you can publish bad images today on any subject

            • joshchandra@midwest.social
              link
              fedilink
              English
              arrow-up
              1
              ·
              edit-2
              21 hours ago

              Right, but I think it’d be harder to get it to unlearn the wrong data if the topic itself is obscure.

              • KeenFlame@feddit.nu
                link
                fedilink
                English
                arrow-up
                1
                ·
                6 hours ago

                Ah. I think I get you, but unfortunately it would probably be a lot easier to unlearn an obscure topic, not the other way around. Poisoning is done at a pixel level if that makes sense