- cross-posted to:
- news@lemmy.world
- cross-posted to:
- news@lemmy.world
cross-posted from: https://lemmy.sdf.org/post/59772642
Image for when they fix it / take it down / rewrite history:
cross-posted from: https://lemmy.sdf.org/post/59772642
Image for when they fix it / take it down / rewrite history:
probability pretty low. Wrong spelling is rare in language models.
At least when they generate text directly. When they generate images of text, that is a different story.
A lot of them are starting to fix that, too. The easy tells are going away fast.
Yes. But it was a massive source of humor for me some months ago when I asked AI chatbots to generate metro map diagrams of my city.
Sometimes the station names they wrote on them somewhat resembled real ones, but were humorously misspelled and in completely incorrect positions and orders. Other times, the overall shape of the network was roughly ok, but the station names were completely nonsensical.
It’s a losing game. Very soon there will be no ways to tell.
True, but we figured out how to deal with the epistemiology in a world with Photoshop; I think we can probably figure it out with generative AI, too.