I'd agree with both takes on that it depends on the meter. Ones which truly approximate password entropy work like you say, however, for some reason, the most common use of such meters is to show how many dartboard requirements you've met while ignoring the actual complexity. When this common approach is used you combine "password must be 8 characters or more" with things like "password must have a number, symbol of ${group}, and capital letter" and the average password complexity is actually made worse for a given length due to pigeonholing.

In the full picture though, in terms of UI/UX, the meter seems like only a downside. In the dartboard use case it's great because it displays what's still needed in terms users work and think with signalling e.g. "you still need a number, otherwise you're all set". People don't really think in bits of entropy though so ll that really is being signaled by either a meter or a normal failed validation hint is "more complexity and/or length needed".

There may be good cases for using a meter while simultaneously implementing good password requirement policy I'm not thinking of though.