9 ms·

Seems like a sensible policy to me. instead of trying to work out if content is AI generated (which isn't possible in general) just check if it's true. not sure

Seems like a sensible policy to me. instead of trying to work out if content is AI generated (which isn't possible in general) just check if it's true. not sure why this is a problem. truth is the only thing that matters in this case

josephcsible 3y ago
Because AI lets people generate answers way faster than anyone can verify if they're true.
- wilg 3y ago
  But how do they verify it's an AI generated answer in order to moderate it? Seems significantly harder than checking the substance.
  - josephcsible 3y ago
    When someone posts 30 long, substantive answers in 20 minutes, you can be sure they're AI-generated.
    - grugagag 3y ago
      They could be posting it on 30 different accounts vpned into different geographical locations. Or 300, or 1000. Eventually the pipeline gets clogged somewhere The quiestion is by inspecting text can one determine if it’s written by a human or llm? Not sure how accurate that would be and we’re back to square spam, now incredibly to eye out as spam but waste good time on broken incorrect information..
      - twelve40 3y ago
        you are describing some weird hypothetical by a very determined person. they are the people who volunteer their time today to moderate, trying to stop dumb copy paste as a first step, be it via looking at the contents or frequency, whatever gets the job done. Always impressed by people opining here on what the world really needs.
    - JdeBP 3y ago
      As explained at length by the moderators themselves, the detection was a lot more multiply-factored than that. They also took into account (for example) massive stylistic changes in posting questions, comments, and answers; because people were only using machines to generate the answers. See https://news.ycombinator.com/item?id=36200852 https://news.ycombinator.com/item?id=36200852 .
    - wilg 3y ago
      Yes, but that's a corner case. What if someone just posts one? And wouldn't you want those substantive answers even if they were AI generated, of course, as long as they were correct!
- fungiblecog 3y ago
  that's an operational issue not a policy issue
- blacksmith_tb 3y ago
  Does SO have APIs you can submit answers to at an unthrottled rate? That doesn't seem like it'd be in their interest to have built - if you limited each account to posting one answer every 30min, that would prevent quite a bit of spamming, seems like?
skepticATX 3y ago
Right now I take a “trust but verify” approach with using StackOverflow. I assume that users are not malicious/incompetent, but verify when it matters. With AI generated answers one must take a “don’t trust but verify” approach. And at that point, why use StackOverflow at all?

josephcsible 3y ago

Because AI lets people generate answers way faster than anyone can verify if they're true.

wilg 3y ago
But how do they verify it's an AI generated answer in order to moderate it? Seems significantly harder than checking the substance.
- josephcsible 3y ago
  When someone posts 30 long, substantive answers in 20 minutes, you can be sure they're AI-generated.
  - grugagag 3y ago
    They could be posting it on 30 different accounts vpned into different geographical locations. Or 300, or 1000. Eventually the pipeline gets clogged somewhere The quiestion is by inspecting text can one determine if it’s written by a human or llm? Not sure how accurate that would be and we’re back to square spam, now incredibly to eye out as spam but waste good time on broken incorrect information..
    - twelve40 3y ago
      you are describing some weird hypothetical by a very determined person. they are the people who volunteer their time today to moderate, trying to stop dumb copy paste as a first step, be it via looking at the contents or frequency, whatever gets the job done. Always impressed by people opining here on what the world really needs.
  - JdeBP 3y ago
    As explained at length by the moderators themselves, the detection was a lot more multiply-factored than that. They also took into account (for example) massive stylistic changes in posting questions, comments, and answers; because people were only using machines to generate the answers. See https://news.ycombinator.com/item?id=36200852 https://news.ycombinator.com/item?id=36200852 .
  - wilg 3y ago
    Yes, but that's a corner case. What if someone just posts one? And wouldn't you want those substantive answers even if they were AI generated, of course, as long as they were correct!
fungiblecog 3y ago
that's an operational issue not a policy issue
blacksmith_tb 3y ago
Does SO have APIs you can submit answers to at an unthrottled rate? That doesn't seem like it'd be in their interest to have built - if you limited each account to posting one answer every 30min, that would prevent quite a bit of spamming, seems like?

skepticATX 3y ago

Right now I take a “trust but verify” approach with using StackOverflow. I assume that users are not malicious/incompetent, but verify when it matters. With AI generated answers one must take a “don’t trust but verify” approach. And at that point, why use StackOverflow at all?