3 Matching Annotations
  1. Mar 2024
    1. But this technology will have somelong-term implications for how we verify claims,

      I can't count the number of times I've already run into issues with this.

      I'll search something and find myself reading something that just seems off—not necessarily of dubious accuracy, but full of oddities like repeating the same or similar ideas several times in the same paragraph, something a human would only do for emphasis, or perhaps to make a more complicated topic more clear but the topic I'm searching information for is something simple like a videogame walkthrough or a recipe. This is merely annoying for inconsequential things like that, but it's a MASSIVE problem that anyone can now buy a domain and use a LLM script to generate an entire site of, say, plausible sounding vaccine denial rhetoric with next to no effort.

    1. Asyou'llsee laterinthisbook,theaverageper-son(wrongly)fallsbackontheirownsenseofplausibility,attemptingtoanswerwhether somethingistrue byaskingwhetheritseemstrue,lookscredible,orfeelsauthoritative.

      I feel like the parenthetical "wrongly" is undue, at least in part, because one does develop a pretty reliable intuition for sketchy resources on the web the more one becomes familiar with the fact-checking methods this book claims to contain. More than that, though because often our bologna detectors DO go off even when we're not quite so trained, and it's wise to listen to that even if it's a false alarm.