

Yeah, email isn’t private, but for me it’s usually that I don’t like my host reading all of my mail to build an profile on me and selling that data. Individual emails in isolation aren’t a big deal, but seeing every email and what company or agency sent it is as problematic as the content even if you encrypt the mail content itself. Emails that I sends I always assume are not private, but that’s a separate issue, IMHO. There’s a lot of private information like what protected classes I am part of, political leanings, places I shop, etc., that can be gathered simply from who sends mail to you and who you send mail to. This is why I self host for most of my email.
That being said I still use gmail as I need a backup option and I use it for things where I don’t want the junk sender to know my domain and spam all of my accounts.
But Proton is really not much more private than Google in several scenarios given their CEO’s stance on several sensitive subjects and willingness to give data on protected classes, journalists, etc., to hostile governments, as an example. They do say they don’t sell your data to ad companies, at least. I don’t know Fastmail at all. And self-hosting is not something I’d recommend if you don’t want to put a lot of time and effort into it. Lots of issues come up like blacklisted VPS IP addresses in addition to the setup itself.
LLMs have uses, some of them are even quite intriguing. But they have to be properly trained. You can’t just throw the whole internet at a baby with very little other training and expect them to not be corrupted by random wrong information. Same goes for LLMs though on a much larger scale. Also, they are often configured to give an answer even when the confidence in it being correct is relatively low. Something an expert would never do, they’d consult only specialized information, not just review the top search results on Google. This is one reason why they “hallucinate”. Commercially trained models just aren’t all that useful as a source of information or to correctly complete tasks. And additionally there are extreme ethical concerns about how it gets the information it’s trained on including using hacking botnets to impersonate a human among other things. A person who’s an expert has to review everything in excruciating detail and so most of the time it’s just more cost effective to just consult an expert in the first place. It’s like going to a proverbial used car car salesman and asking how cars work. Sure they might have picked up a fair amount of information from being around mechanics, but some of it is wrong and what they don’t know they’ll just make something up that sounds mostly plausible.