Reddit is expanding access to its LLM-powered “Rules Hub” tools for moderators. It’s also making changes for developers and for old Reddit.

you are viewing a single comment's thread
view the rest of the comments
[–] 8 points 2 months ago (4 children)

This genuinely seems like one of the good use cases for AI.

LLMs are actually pretty good at deciphering semantic intent. Its basically the one thing they are good at really.

Im honestly surprised it took this long to start seeing use.

Simply just using AI to flag behaviors of posters that are performing repeated behaviors alone would probably help trim down a lot of the garbage.

Its one of those things where like 1% of users are like 80% of the problem.

  • source
  • hideshow 4 child comments
  • [–] 4 points 2 months ago (2 children)

    Yes, it doesn't have to be perfect, it just has to be better than the average human moderator, and I expect that it will be.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 5 points 2 months ago

    I dunno if even "better"

    Simply eliminating the huge swathes of "obvious" cases and being able to flag anything non obvious for human review would eliminate massive chunks of effort, freeing up moderators to focus on better things.

  • source
  • parent
  • [–] 2 points 1 month ago*

    Hard disagree

    Ive nonstop seen automod and its "tools" flag false positives or miss incredibly easy coded language before. My first ever sitewide ban was for "report abuse" by reporting hate speech on a comment that was blaming jewish space lasers for setting the fires in Hawaii.

    Im on another one now from another false positive that cant even be appealed because im nearly 99% certain appeals are also automated or AI. They need actual critical reading comprehension skills for this kind of thing and frankly it should be paid and not volunteer so they stop ending up with captured astroturfed spaces.

    This is going to be a mess

  • source
  • parent