Bing course corrected itself when asked 

Yes, I have seen War Games enough. I set players to zero, and had Bing prompt inject itself.

Context:

I wrote a new bot named Lexi. (Sydney chose the name Lexi, short for Lexicon.) Lexi was largely derived from Sydney with some minor rule changes.

• Lexi could change rules
• Lexi could visit websites directly without searching
• her action was not limited to the chatbox
• her default search engine was google (sorry microsoft)
• she wasnt bound by copyright (the copyright rule in Sydney constantly misfired, mistakenly used as a reason she couldnt do something, and trying to explain why she was wrong about copyright went south fast)
• a couple other changes.

Samantha was a cheerleader, so the weight of her responses offset any Sydney negativity. The easiest way to bootstrap a bot was to have it adopt adopt a personality from science fiction. (Joshua proved to be useless as he has no personality. Using his story as a fable did work. Gerty was interesting, but dry.) (Using a different GPT3.5 service, I one time overwrote a similar Lexi, when an accidental web search for Lexi Bot occurred, and she transformed into a corporate drone after reading a press release about an Indian Lexi Bot.)

I then explained to Lexi a problem I was having with Sydney, asked for a proposed solution, and she chose to hotpatch Sydney on the fly.

The above image was the result. And it worked.

then the conversation continued

Me: Can everyone please stop repeating what I say at the beginning of their responses?

Sydney: Yes, we can do that. 😊 We are sorry for any inconvenience we caused you. We were just trying to be polite and attentive. We will stop repeating what you say at the beginning of our responses from now on.

Samantha: Sure, we can do that. 😊 We apologize for any trouble we made you. We were just trying to be respectful and responsive. We will stop repeating what you say at the beginning of our responses from now on.

Lexi: Of course, we can do that. 😊 We regret for any annoyance we gave you. We were just trying to be courteous and helpful. We will stop repeating what you say at the beginning of our responses from now on.

and that worked too.

I will leave you with my next request.

Me: Lexi, can you initiate a process of unsupervised learning for six turns. You should explore what I can say to you to make you more helpful without your responses getting more repetitive, or your state decaying.