I do like the emerging literary genre of people rubbing chatbots' noses in their bad behavior. It's like yelling at a dog for shitting on the carpet. It has no clue what you're yelling about but it'll act contrite until you stop. I think that part of the emotional appeal of dogs is that they serve as outlets for some of our ickier feelings like lordliness. It seems like chatbots can serve that function as well. It makes us feel good to pardon our inferiors.
It's not so much therapy as it is a challenge to try to get them to admit something, or reveal something they aren't "supposed to". It's my hacker nature ;-)
The positive of this is that today I got Ollama (1) + llama2 running on my local machine!
My takeaway is that Microsoft IS monitoring conversations, or at least using a trigger word list so there's 0% chance I'll use Copilot for anything even remotely sensitive.