• christian [he/him, any]@hexbear.net
    link
    fedilink
    English
    arrow-up
    2
    ·
    26 days ago

    It seems like it would need very direct rules in the code to just defer to a human tech in the event of not “knowing” the answer.

    That would require a wholly different technology with some ability to interpret the things it’s saying and assess their validity. It’s a lot more cost efficient to have your AI spew bullshit and do damage control afterwards.

    • dRLY [none/use name]@hexbear.net
      link
      fedilink
      English
      arrow-up
      2
      ·
      25 days ago

      Well then, it seems that it would be good motivation if more people are able to find ways to force the bots to give deep discounts on things. Would need to be clever in tricking the AI so that it isn’t outright obvious in the prompts to aid in avoiding the companies just accusing the buyers of hacking.