I remember a conversation I had with a coworker back in Mountain View, before the dotcom bubble, around when everyone was putting their appliances on webcams, and how it could feasibly be used to map peoples' habits and violate their privacy for life if it they were put in workplace breakrooms...

Then we all used Zoom all day and night during COVID, and it is all stored for at least six months. It isn't a big leap to jump from user experience research for AI to that.

I fail to see how caution about anything that can rapidly gain intelligence and lock out its perceived creators is paranoia.

I mean it is neat. I work with it. I have diddled with LLM since 2010/2011. That does not mean I have not seen people make stupid mistakes and confuse right and wrong constantly. So why do we think our models can discern it?