“Fairly” carries a lot in that sentence. From the polling thread:
Systemic errors of three or more are pretty common and state polls off by three even more so. That is fairly accurate for some value of fairly maybe but it is also blunt.
The point is made in the NYT polls tracker section nearly as a sticky:
In such a close race, even the slightest movement in the polls takes on outsize significance. For that same reason, even a modest error in the polls could yield a very different result. If the polls underestimate Trump yet again, even by a hair, he would easily win the Northern battleground states and therefore the presidency. But the opposite could be true: Polls underestimated Democrats in 2022, and Harris could win easily if they do that again, even slightly.
It cannot be emphasized enough: if there is a historically typical systemic error then this election ends up not being close electorally; she could win fairly handily or lose humiliatingly. A fairly typical systemic error in her favor has her winning the popular vote by nearly 6 and even coming close in FL. Of course if it goes the other way she loses the popular vote and is pretty crushed in the EC.
Polls are good, not broken, and too blunt of a tool to discriminate between those dramatically different outcomes.