• mossy_
    link
    fedilink
    English
    arrow-up
    51
    ·
    6 days ago

    in hindsight, believing robots would be compulsively honest was a bit optimistic

    • mkwt
      link
      fedilink
      English
      arrow-up
      26
      ·
      6 days ago

      They’re not really being dishonest, per se. They’re just bullshitters spouting bullshit. They don’t care about true or false. Because the truth does not appear in the objective functions they are optimizing.

      • Swedneck@discuss.tchncs.de
        link
        fedilink
        English
        arrow-up
        2
        ·
        4 days ago

        i think that gives them too much credit, even. They don’t know what is true or false, LLMs can’t even really properly do logic in the way that normal programs can, they just have to emulate logic via really fancy statistics and the fact that the things they’re trained on contain responses to logic.

    • Telodzrum
      link
      fedilink
      English
      arrow-up
      1
      arrow-down
      1
      ·
      6 days ago

      You can make them honest or you can make them obsequious. Everyone who works in the tech industry wants nothing more than to told they’re smarter and better than an “average person.” So, this is where tech bros, code monkeys, and everyone else in that mistake of a field will drive the technology.