This is why I mostly look at their polling averages and not RealClearPolitics or some other basic aggregator.
As he notes their pollster ratings incorporate not only past performance, but methodological transparency. In other words if a pollster shares how the sausage is made or not. As you can figure a lot of the Trump biased pollsters aren't very transparent. A lot of the household names like Quinnipiac or Pew are.
Low quality pollsters will not influence the averages that much. High quality pollsters will.
And the averages make adjustments for house effects (bias). This goes both ways. There are Dem biased pollsters as well as Trump biased pollsters.
And the averages account for "flooding the zone". Especially with some of the newer online pollsters, they are able to release polls very often. The averages penalize for a lot of polls being released in a short amount of time, so one firm doesn't crowd out the averages.
As he notes, incorporating low quality and biased pollsters with adjustments is better than excluding them historically. No pollster is perfect and herding can happen as well. Much like publication bias in research, you get a better picture usually by incorporating all the info.
Of course, FiveThirtyEight does blacklist some pollsters that violate their ethics guidelines. Rasmussen Reports has been blacklisted. Along with some other firms you probably have never heard of.