Skip to content

Detecting spam content that is not overtly so

Uncategorized
  • In another thread, @puppygirlhornypost2@transfem.social talks about challenges regarding proliferation of spam (or other unwanted content) on the fediverse:

    > ... I was talking about reddit's "posts require approval" mode. Something that might be of use for new accounts registering on pixelfed instances to prevent things like gorespam from occurring.

    NodeBB has the post queue (our name for it), and it works really well to stop drive-by spammers. Most cheap spam is easily identified as such, and it's only a couple clicks to send them to /dev/null.

    More recently, and especially since we've been federating posts outward, we've started encountering possibly LLM generated content entering our post queue. It could be that these accounts are legitimate users creating an account to respond to remote content, but it's equally likely that they're attempting to create plausible content in order to pass validation, before proceeding to post spam.

    At least here on community.nodebb.org it's trivial to discover, because this forum is for NodeBB support/discussion, and a post out of the blue responding to a post about Trump or Israel/Palestine is very obviously not meant to be on this forum, although it's definitely one of those "hard problems" for more general instances.