With the right series of prompts and criteria ChatGPT could probably whip up an effective virtual nanny.
It seems plausible to add penalties of some sort that make me take it seriously but hard for me to align it in a way that handles edge cases well