Tweets like this are doing big numbers at the moment, and while its worth underlining that Trump (a) can still win and (b) has an Electoral College advantage, its also worth remembering that polling methods have changed substantially, particularly in state level polls, after 2016 https://twitter.com/WalkerBragman/status/1300122482528129028

One reason state level polling, in particular, looked so favourable for Clinton at this point is that pollsters in 2016 did not weight by education. Their samples had too many white graduates (who leaned Clinton) and not enough white non-graduates (who leaned heavily Trump)

Polls in 2020 nearly always do weight by education, which *should* make them more accurate (of course other sources of error are possible). But this means comparison with state level polls in 2016 are comparisons of apples with oranges

Another thing to bear in mind is that many of the states Trump won in 2016 were severely under-polled, meaning the RCP "averages" in places like WI or MI are often based on one or two out of date polls. They are polled much more intensely this year.

For example, in Wisconsin there were only two polls by one company driving the RCP average for the whole of July and August 2016.
In 2020, there are 10 polls from 6 different companies

One final thing - the margins are important, but so are the levels of vote. It is easier to close a 5 point gap when 20% of voters are undecided/backing a third party (as in 2016) than when the figure is 10% or below (as is typical in 2020).