That only helps if you know you're not discussing anything sensitive. The problem is, people often do discuss sensitive matters. I don't think I even need to quote Cardinal Richelieu here, this is hacker news, no need to preach to the choir.

People can't threat model because they don't really know what the governments consider interesting and a threat to their exercise of power.

So always use E2EE.

I think the point is that they're upfront about it, so you can treat the communication as though it were public (which is a pretty good default assumption for communicating on the internet in general).

I'm one of those people that refuses to accept that "privacy is dead"; but there's also a lot of casual/low-stakes communication that I treat like a personal conversation in an IRL public setting, operating on the assumption that a stranger can/will overhear it.

I just explained, people don't know how to threat model. An average Joe doesn't think "Whoa Ok, so this is TLS only, that means I should assume the server has access to content that might have implications depending on policy, level of government collaboration, and that foreign actors might also have access if they hack the server. Therefore, I should evaluate in real time my conversation, constantly thinking how an outsider might perceive it, and if the other peer starts talking about something private, pack my bags and move abroad".

>I'm one of those people that refuses to accept that "privacy is dead

I never claimed it was. I was making the point that you should protect everything so you don't leak metadata about when you're having private conversation. That's exactly what happens when you e.g. enable secret chats in Telegram. You're telling the company "I'm now talking to Bob, and I'm intentionally making the decision to not share that data with YOU". That's really, really valuable metadata to governments.