I bet these polls are within a couple points of the reality, as they pretty much always have been (even in the era of cell phones). They specifically weight the results based on the expected non-response of various groups. They account for the most obvious objection that anybody could raise about a modern poll (this one).
The polls in 2016 were largely on target. Clinton won the popular vote by roughly the predicted margin and the few key swing states that lost her the race had results that were largely within the margin of error. Lots of people took 2016 as an indication that polling is no longer good. That’s the wrong lesson to learn from 2016; it just doesn’t match the facts.