The funniest argument I've seen is that the polling has changed and adapted since 2016, but that's truly hard to believe. Polling is about your sample, and it's not likely that any of these pollsters have been able to change their sample to include the Trump voters they missed in 2016. Now models/projections should have been fixed and should be able to account for this, but considering the polls look so similar to 2016 and the models are still so drastically in Bidens favor it's hard to imagine even the models have changed. (which is weird to me)
But I ran some numbers, if the same shift happens, I took the polls now and scewed them to what they would be if the same swing happens this year (polls on Oct. 23, 2016 versus the actual results)
This is where the election is won or lost:
Georgia (Trump +0.3)
Pennsylvannia (Trump +1.8)
Arizona (Trump +2.4)
Wisconsin (Trump +3.1)
Biden would need 3/4 of these states to win.
Here are the full predictions in battleground states if anyone is interested.