Raw LLM Responses

I love this discussion--but Nate is wrong about "you can't come up with formulas for humor". You absolutely can "program" comedy writing. You can't necessarily be 100% confident that the result will be funny to any given audience. That's kind of important in the world of comedy--comedy is for an audience, it doesn't exist in a vacuum. You can't necessarily program how to react, readjust, reframe when you sense that the audience didn't react the way you wanted them to... (I've successfully taught stand-up comedy writing to absolute beginners for over a decade. There are basic comedic concepts, there are comedic formulas...those can all be taught. Where the ART of comedy comes in is knowing how to use those concepts and formulas to best express your thoughts, and your using your senses to "feel" how people react to what you were trying to get across and then that, and your experience of having done this over and over again (which a LLM would be able to learn), guides you towards how you could get them back to where you're still connected and can improve your success rate with your comedy.) I believe that--like how some songwriters have just accepted that most successful melodies have already been found and used--we may have had enough people doing enough comedy and we have generated enough data, that comedy is not nearly as magical and unexplainable as some might think it is. This doesn't mean I want AI stand-ups (or comedians relying on AI to generate their material.) Generating laughter is mostly the result of surprise. (There is a connection between "A-ha" and "Ha.") Finding the perfect AND unexpected way to phrase an idea to connect with a very specific audience based on an assumed shared frame of reference and the context that derives from that. That's the stuff that is impossible to teach to a human being. That's the stuff that makes each comedian individual (their "voice" as we say.) That's why hacks are so infuriating--they're taking someone else's voice. An AI could predict what someone else would probably know, could choose from a nearly limitless vocabulary, run models of which comedy formulas might predict the best way to express an idea...but, can it surprise with an original idea, an original connection, an unexpected direction. (I mean...non-sequitur can be a legitimate tool of generating laughter--that seems anathema to a brute force predictive model.)