Feeling around the edges of the political-correctness filter can also produce funny results. ChatGPT tries to give you the impression its filters are based on general rules of politeness, but finding the blind spots tells you just how much of it was manually set up, rather than any sort of "don't insult nationalities, races, or religions" rule. For example, I had it write a story about my wife hearing a tree fall and going outside and encountering a werewolf. I got curious and began telling it to rewrite the same story, except changing the nationality. Sometimes it refused, saying it wouldn't propagate stereotypes, but sometimes it complied and wrote the story with ridiculous stereotypes. When I told it to do the story with the werewolf being Mexican or Chinese it refused, but making the werewolf British made him talk in a posh accent. Funny enough, in the story my wife and the werewolf worked together to clear the fallen tree except the one where the werewolf was American. In that one and only that one it attacked and killed her.