Login
You're viewing the front-end.social public feed.
  • Nov 2, 2026, 1:19 PM

    Fucking great. Sigh. For me, Seeing AI by Microsoft is now describing images that I know are human photographed to be AI generated. I wish there AI describing feature acted more like Ira and just describes the picture not tries to interpret if a thing is AI generated or not. #blindness #Blind #AI

    💬 1🔄 6⭐ 0

Replies

  • Nov 2, 2026, 2:27 PM

    @WeirdWriter as a visually impaired blind photographer:
    1. I use alt text and add even more detail and context in the captions - so should everybody else
    2. [intense eye twitch]

    💬 1🔄 0⭐ 0
  • Nov 2, 2026, 2:31 PM

    Indeed! I predict the companies that make these cloud LLMs will next try and sell us the solution to these incorrectly interpreted assumptions. Here we go! Local models can't happen fast enough! @MostlyBlindGamer

    💬 1🔄 0⭐ 0
  • Nov 2, 2026, 2:44 PM

    @WeirdWriter they’re just as likely to get all of it wrong. There are already competing technologies for proving image authenticity, by the way - deterministic ones that start at the source.
    Provenance and description have the same problem: they only hold as a through line from original creation.
    An evidence bag at a police department that’s lost its original context can either be a bad Halloween party or world-level-movie-star-divorce event.

    💬 1🔄 0⭐ 0
  • Nov 2, 2026, 4:28 PM

    why couldn't we have *more deep learning* and *more deterministic models* that could at least help along with the image descriptions? Probably because they don't make anybody money and one can't say something is for *everything* when it is especially targetted, ugh! @MostlyBlindGamer

    💬 0🔄 0⭐ 0