When German journalist Martin Bernklautyped his name and location into Microsoft’s Copilot to see how his articles would be picked up by the chatbot, the answers horrified him. Copilot’s results asserted that Bernklau was an escapee from a psychiatric institution, a convicted child abuser, and a conman preying on widowers. For years, Bernklau had served as a courts reporter and the AI chatbot had falsely blamed him for the crimes whose trials he had covered.

The accusations against Bernklau weren’t true, of course, and are examples of generative AI’s “hallucinations.” These are inaccurate or nonsensical responses to a prompt provided by the user, and they’re alarmingly common. Anyone attempting to use AI should always proceed with great caution, because information from such systems needs validation and verification by humans before it can be trusted.

But why did Copilot hallucinate these terrible and false accusations?

  • Terrasque
    link
    fedilink
    English
    arrow-up
    1
    ·
    edit-2
    2 months ago

    randomly sampled.

    Semi-randomly. There’s a lot of sampling strategies. For example temperature, top-K, top-p, min-p, mirostat, repetition penalty, greedy…

    • futatorius@lemm.ee
      link
      fedilink
      English
      arrow-up
      1
      ·
      2 months ago

      Semi-randomly

      A more correct term is constrained randomness. You’re still looking at probability distribution functions, but they’re more complex than just a throw of the dice.

    • vrighter@discuss.tchncs.de
      link
      fedilink
      English
      arrow-up
      1
      ·
      2 months ago

      randomly doesn’t mean equiprobable. If you’re sampling a probability distribution, it’s random. Temperature 0 is never used, otherwise a lot of stuff would consistently hallucinate the exact same thing

      • Terrasque
        link
        fedilink
        English
        arrow-up
        1
        ·
        2 months ago

        Temperature 0 is never used

        It is in some cases, where you want a deterministic / “best” response. Seen it used in benchmarks, or when doing some “Is this comment X?” where X is positive, negative, spam, and so on. You don’t want the model to get creative there, but rather answer consistently and always the most likely path.