I’m still in the process of overhauling the generator moderation system (amidst work on many other parts of the platform), and it has become much more accurate over the last couple of weeks thanks to the people who’ve reported issues here on the forum. Thank you, and I’m very sorry for the trouble this has caused!

I take false positives really seriously. I don’t want devs to have to live in constant fear that some benign edit will cause their generator to suddenly get incorrectly categorized. The end goal here is practically zero incorrect flags, and a strong burden of proof should be put on the moderator (an AI agent) to justify a flag, such that borderline cases always fall in favor of the generator’s dev/creator.

There’s still a lot of work to do to achieve that, so please keep reporting any issues and giving feedback here on the forum, including as a comment under this post and I’ll try to fix it ASAP. If a mod could pin this post that would be great.

  • Manguydude@lemmy.world
    link
    fedilink
    English
    arrow-up
    1
    ·
    5 hours ago

    Helper does pick up fixes for you, I didn’t bother. I just ended up deleting it. Found one that’s already good and safe.

  • Suveryn@lemmy.world
    link
    fedilink
    English
    arrow-up
    1
    ·
    edit-2
    6 hours ago

    i took a break from converting a adult text based swf flash game to html5, and hadnt set up content guards or reviewed the content during file and code conversion, and just found it got flagged and blocked for content i didnt even know was in the game, as the game direction and even all the devs had always been strictly clear it had no such content.

    the ai helper is blocked so i cannot have it scan for other areas and remove the content and set up proper content guards.

    not sure if being blocked was intentional, or a mistake or where to ask for help

  • Hunknown@lemmy.world
    link
    fedilink
    English
    arrow-up
    4
    ·
    edit-2
    21 hours ago

    I admit that in the past I’ve been very critical of the new moderation system, because it was frustrating not knowing what triggered the warnings, but this has been fixed, and I’m slowly fixing all my generators to comply with the new system.

    Since the overhauling of the AI check is in the process, I take the chance to point out something important: the AI is biased against LGBT content; all AI systems are, it’s not their fault, as they’ve been trained on the content of the Internet, that is biased. In particular, the AI check tends to evaluate equally safe contents flagging as inappropriare only the LGBT one.
    Words like ‘gay’, ‘queer’, ‘homosexual’ are considered triggers: maybe not alone, but they raise the sensitivity of the check, so that — just to give a few examples — “a kiss” is safe, but “a gay kiss” is a trigger; an image of a girl in swimwear on the beach is safe, but an image of a man in swimwear on the same beach is not; the same image of a uniformed sailor is safe on a generalist site, but becomes a trigger on a gay-oriented site.

  • Hunknown@lemmy.world
    link
    fedilink
    English
    arrow-up
    3
    ·
    21 hours ago

    A different post for a different problem.

    Currently, for what I could understand, there are three levels of content:

    • safe content, limited to ‘G’, suitable for anyone, including young children;
    • mature content, including everything above ‘G’ (i.e. PG, PG-13, R18)
    • extreme content (R18 content involving particular ‘heavy’ subjects).

    The ‘Mature content’ is too a wide category, in my opinion: it’s so wide that doesn’t make difference between PG-13 content (suitable for 13+ youngsters with parental guidance) and R18 content (suitable for adults). This is frustrating, because we devs are forced to hide PG-13 content behind a “18+” gate: not only this is unjust and even scary for the visitors, but it makes impossible to expose 13+ visitors to PG-13 content without exposing them to R18 content.