• Communist@beehaw.org
      link
      fedilink
      arrow-up
      4
      ·
      1 year ago

      It’s not, this method of analysis is terrible, they’re just asking gpt4 to grade the responses, not actually testing anything beyond that.

      • aponigricon@beehaw.org
        link
        fedilink
        English
        arrow-up
        4
        ·
        1 year ago

        That doesn’t necessarily invalidate the point they’re making. Other forms of analysis, strikingly, provide pretty much completely equivalent results.