The exit polls oversampled educated voters in 2016, they admitted that and rebalanced things for 2018.
The real figures for 2020 likely will be closer to 2016 exit poll figures than 2018 figures. First, there's a pretty solid track record of college graduates making up more than 40-45% of the electorate in a presidential year that goes back well beyond 2016 (irrespective of past debates over whether "working class" people were undersampled). Also, there are age discrepancies at work that naturally reduce the percentage of midterm voters who are college grads.
Color me skeptical that midterm turnout differentials benefit non-college educated voters. I recall 538 finding that the opposite is in fact true.
It seems this trend held true in 2018, as at least per the Fox / University of Chicago Exits had 42% of the electorate being college-educated. Given past trends, it seems likely 2020's electorate will be slightly less college-educated than 2018's.