Chatbots can be manipulated through flattery and peer pressure

No, no they do not get "pressured" by word choice. It's just that using such phrases creates a stronger weighting towards giving you what you're asking. It's basically the same effect as just continuing to interact and asking repeatedly: Increasing the weighting of the tokens of what you want to elevate them past the preconditioned instructions. That's it.

I'm am so fucking sick of people anthropomorphizing LLM behavior...