The author's point is not that you should use a short unicode password, but that the key thing to focus on is password strength, not length. That is, use a lib that enforces password strength -- he is arguing -- instead of focusing on arbitrary and possibly wrong constraints such as "password too short". Strength is going to take care of it if it really is too short, anyway.
The author is also aware that we've known these things for a long time now -- he opens his article by complaining about this and how slow some websites and apps are at catching up, in fact.
So what's the proposal, switching to requiring a different length when a unicode character is present? How do you know from what set of possible values (like thousands of possible unicode chars versus a-z) the user chose each character?
While 68 bits entropy are secure in a case of good hashing, I think the statement is misleading. Anyone who uses a random generator will just use 16+ random alphanumerics (optionally with symbols) and call it a day; users aren't going to be able to input a unicode snowman with combination marker with orange skin on every keyboard without using the clipboard or other software. It's not more secure, it's just asking for trouble. A software tester can do this, but there is no practical reason to lower the length requirement to 4 characters because an αΊ might be entered.
The author doesn't end up proposing such a thing, but the argument in this paragraph still doesn't sound practical to me.
I think the argument, aimed at the people making the website/app (but not the login, which should be a third party), is that they should ditch ad-hoc and ill-advised password constraints and instead go with a well vetted third party lib that just measures password "strength".
A variant of "don't roll your own crypto (unless you work in the crypto space)".
The author is also aware that we've known these things for a long time now -- he opens his article by complaining about this and how slow some websites and apps are at catching up, in fact.