4

When asked a question it provided wrong answer.

It did give correct list from April 13 until April 28, which is 6 days. But in the end it calculated 8 from somewhere.

  • Vlad replied to this.

  • Vlad Thinking about it, it might not be clear enough. Knowledge of the limitations of LLM (e.g. plausible but hallucinated outputs, bias, sandbagging, logic & math) is not widespread, and there is no information about this in the Kagi documentation.

    Would be nice to have a link in the tool tip, that links to a Kagi documentation page explaining general limitations of LLMs or something like that.

    At the moment it seems that Kagi users are mostly tech savy, but when Kagi grows its userbase, more & more "normies" will flock in. Managing user expectations will then become crucial.

      azdanov LLMs are not good at doing math, is the objection here that you as a user do not have correctly managed expectations with this feature?

          I guess so. If I didn't expand the answer and didn't see all the days listed, then I'd get the wrong answer 8 shown above the fold.

          Could be a bit dangerous, since I might not know that quick answer can provide incorrect answers.

          • Vlad replied to this.

            Vlad Thinking about it, it might not be clear enough. Knowledge of the limitations of LLM (e.g. plausible but hallucinated outputs, bias, sandbagging, logic & math) is not widespread, and there is no information about this in the Kagi documentation.

            Would be nice to have a link in the tool tip, that links to a Kagi documentation page explaining general limitations of LLMs or something like that.

            At the moment it seems that Kagi users are mostly tech savy, but when Kagi grows its userbase, more & more "normies" will flock in. Managing user expectations will then become crucial.

              12 days later
              No one is typing