• nightlily@leminal.space
    link
    fedilink
    English
    arrow-up
    5
    ·
    edit-2
    3 days ago

    If it were that easy, we wouldn’t be in this situation. Education wouldn’t be under attack, social media wouldn’t be flooded with bots. LLM detection is incredibly unreliable and anyone saying they’ve cracked it is selling snake oil. There are techniques for image diffusion that are holding up currently but text is another story.

    Checking a checkbox again relies on these chucklefucks being honest and decent and respectful people which they fundamentally aren’t.

    • wonderingwanderer@sopuli.xyz
      link
      fedilink
      English
      arrow-up
      1
      ·
      3 days ago

      It could be as simple as creating a whitelist of verified contributors who are trusted and routing everyone else through a filter. Or identifying the most serious offenders by looking at volume and frequency of contributions, making a list of the ones that are obviously automating, and routing those through the filter, which should still significantly reduce the number that require human review.

      Neither of which are perfect solutions, but it could make the situation more manageable if you don’t trust people to be honest about whether their info was generated with AI.