This is the same argument as "self-signed certificates are useless". They aren't. Not all interception is MitM. With spy agencies operating passive taps on infrastructure, as a fraction of your intercepted traffic, virtually none of it us. Completely unsigned, unverified encryption is still a massive roadblock to dragnet surveillance.

>does literally nothing to prevent MitM attacks

It does more than nothing. Especially if you make the reasonable assumptions that 1) the user is not being MitMed most of the time and 2) the user is likely to have connected to the (small) set of sensitive sites before they experience their first MitM.

Example: my home internet is unadulterated. I check my mail. My computer saves the certificate. I go to a coffee shop. I check my mail again. Somone tries to MitM me. My computer instantly spots a changed certificate.

Obviously we should still use and verify certificates. But I notice a nasty trend of making the perfect the enemy of the good in security. There's a perception that if it's not perfect, it's worse than useless and should be shunned (probably due to historical bad experiences with false senses of security).