Bryan Han, a programmer who often posts comics at reddit, experimented in outsourcing humor. He asked ChatGPT for an extremely funny idea for a comic, and then he illustrated what the algorithm gave him.
Are you laughing? If not, is it because you are underwhelmed or just confused? The discussion about this comic turned to speculation as to why ChapGPT thought this was "extremely funny." Maybe it's an anti-joke, which is funny specifically because of how it's not funny. In other words, a failure of a joke. But maybe there's some way that the punch line resonates with artificial intelligence, as if a neural network is constantly striving to be relevant and valuable. Or maybe it was just fed too many anti-jokes. Then again, the algorithm may have seen this in a higher level of humor, in the profound angst of anyone who makes an honest mistake. Or maybe it was a simple confounding of expectations that filled the network's definition of a punch line. Han tried to wring humor out ChatGPT a second time and got another four-panel comic.
In this case, we see that it must be the rule of making a punch line by subverting expectations that is guiding the algorithm. It's a least a bit funny this time.
Redditor jerog1 prompted ChatGPT to produce dialogue for a four-panel comic and then ran a few of those ideas through Midjourney to illustrate them. You can see in this sub-thread that Midjourney doesn't do well with words, which is okay because they weren't funny anyway. -via Geeks Are Sexy