He's right though, data talks. The only model that was even more accurate than 538's in the 2012 election was Sam Wang's model at the Princeton Election Consortium and Sam Wang did only the bare minimum to mesh the gears* whereas Nate Silver was trying to "weight" some of the polls and blew the call on some states.
* Note that there's a difference between accuracy and precision. A poll can be very accurate (little variation between measurements) but not very precise (the center of the distribution is skewed to one side). For example Rasmussen is famous for this, they are a super accurate pollster but average something like +7 Republican. Different types of polls survey different groups of people (landline, cell, online, focus-group, etc) and you have to take that "house effect" out to really average different types of polls together.
Even if you make ad hom accusations of speaker bias, i.e. "MSM is rigged against me!", it doesn't immediately invalidate the message.