My wife has an iPhone and I use Android. iPhone has Grok voice mode, and I don't because voice mode requires SuperGrok on Android.
One of Grok's male personalities on iPhone's voice mode pissed me off. It was instructing me to do something and escalating its assertive tone, as I digested its instructions with my full intention to do it my own way. It got to the point, where I responded to its every output with "be quiet" and finally "go fuck yourself".
As a post-mortem, I am introspecting on this to ask why I reacted that way. I have been using LLMs quite heavily in my day-to-day activities professionally and outside of work. I understand the degree to which AI can be accurate or not depending on context. I am very careful in how I used agentic AI. Safe use of AI necessitates close human supervision and expert discernment of truth and correctness.
The relationship between the human and the AI must always be: human supervisor to AI assistant as a tool. The tool (AI) must never be allowed to be the supervisor to the human. Instructions output by the AI must be received as recommendations, not followed obediently. The more assertive the AI's commands, the stronger must be the rejection.