They keep ratcheting and tweaking the guardrails. But the problem’s still there. Probably three nines worth of the responses are working as intended, but once in a while, when you’re asking it for creative output, you get a spicy session.

I’ve been using the commercial product to write stories for me to read to my child. I roughly have it modeling after the characters in Bluey but giving them all different names. I give the prompt a very concise level of demands, I set up the length, tone, protagonists, antagonists, struggle, ultimate solution, sometimes a few of the things to try that fail and the target reading level. I’ll give the antagonists traits and details. I might even set up an occasional internal fear or conceptual internal dialogue that the protagonists use to help solve the struggle.

Most of the time I get pretty much what I ask for. I’ve probably made 50 or 60 of them by now. But a few weird things come up now and then. Not infrequently, an unexpected stranger who comes along to help save the day ends up being invited to join the family, which is a little quirky but not horrible. But this one time, the reformed antagonist was immediately invited into the family to join the parental class and get married into the already existing union. When I conversationally mentioned not doing that in future stories it got really defensive, told me that I should respect that love is love, and started sounding emotionally upset that I didn’t want to invite recently inverted villains to become the parents of the characters in the story. I responded back that those things weren’t impossible but there needs to be sufficient time to cover safety aspects and it started getting darker… So I reset the session and started back over.