My first prompt more basic than yours: “generate 10 red flags for characters as writing prompts.” It gave me a lot of supernatural ones (like yours), which wasn’t the vibe I was going for, so my next move was “quirkier, but with no magic.” That’s when it started generating ableist responses, so I said “can you make a list of red flags that does not include symptoms of mental illness? many of these imply that having OCD or autism is a red flag, which is inappropriate”. I argued with it a bit about autism traits being called red flags, and finally gave it specific examples of what I was after: “What I want are silly red flags that might turn people off a character, like ‘not owning a single book’ or ‘pees with the door open’.” It did another couple rounds of autism traits, so I pivoted to “generate silly/quirky ‘red flags’ as if for a dating profile, without using any symptoms of autism, OCD, or mental illness.” The dating profile caveat didn’t help at all, and produced some really weird results (many of which were actually violent / scary). I then added an example from one of its previous generation attempts: “here’s an example: The character eats everything with their hands, including soup.” It gave me a few okay ones from that, and by that point I’d collected enough to use in my post, but really only got maybe one or two good ones from each attempt, if any. There’s definitely a learning curve to how to prompt ChatGPT and maybe I’m too conversational (I left a lot of the convo out, because it was just me patiently explaining over and over that having autistic traits is not a red flag) but there’s definitely also a problem with how the program interprets those prompts.