The author seem to be missing one of the big reasons ridiculously low TTLs are used: it lets passive eavesdroppers discover a good approximation of your browsing history. Passive logging of HTTP has (fortunately) been hindered as most traffic moved to HTTPS, but DNS is still plaintext.
Low TTLs mean a new DNS request happens (apx) every time someone clicks a link. Seeing which domain names someone is interacting with every 60s (or less!) is enough to build a very detailed pattern-of-life[1]. Remember, it's probably not just one domain name per click; the set of domain names that are requested to fetch the js/css/images/etc for each page can easily fingerprint specific activities within a domain.
Yes, TTLs need to have some kind of saner minimum. Even more important is moving to an encrypted protocol. Unfortunately DOH doesn't solve this problem[2]; it just moves the passive eavesdropping problem to a different upstream server (e.g. Cloudflare). The real solution is an encrypted protocol that allows everyone to do the recursive resolution locally[3].
[1] https://en.wikipedia.org/wiki/Pattern-of-life_analysis