For me, genLLM is great for tidying up text and cleaning up grammatical errors, or for generating template material that has no specific style, but as a source of truth (ie: "What were the Paris accords about?"), it is VERY dangerous to use because it has no real intelligence - rather it's just aggregating existing information and cannot discriminate between probably factual and probably fictional.
I worked on systems like this in 1990s (yes, it's really not a new thing, sorry) and yeah, we were surprised at how well it could synthesise natural output, but also that it was equally good at generating contextually meaningful output as it was at generating utter gibberish.
In the end, I've found that it just gets in the way and can lull you into a false sense of security exactly because what it generates can feel so natural and correct.
Basically, we created the world's best con artists.
Like others here, I have no problem with it being on my system - but I have a big problem with it being forced on me against my will with no way to switch it off.
Case in point, as I type this, Edge is constantly predicting what I'm going to type next and show it on the screen. 90% of the time, it's wrong and I've learned to ignore it. But sometimes it interferes with the text edit box on some websites (Twitter being a notable case) and it makes chunks of what I've typed vanish - endings of words most often, but sometimes it just goes nuts and starts type backwards or just types gibberish.
And I can't turn it off.