In 2016 the polls got it so wrong and didn't know why; this time around it's even worse, here's why:

The polls this election are invalid, not due to any fault of their own, I think there is no selection bias, I think the sample sizes are fine, but consider this:

It's the 1950s and you poll, without error, proper sample size, proper selection distribution, and you are asking people if they are gay or not—you will find that <0.1% of the population is gay. This poll will have been done perfectly as far as statistics are concerned, yet the results will be WAY off (we now know that number is >5%). The problem isn't with the polling, the problem is nobody wanted to publicly acknowledge they were gay in the 1950s for fear of societal consequence.

I've talked to close friends on both sides of the political spectrum and if there is one thing I've learned it's that some, a statistically significant amount, of Trump supporters in 2020 are "in the closet". They don't want to publicly declare this for fear of societal consequence.

For the same reason the gay community was under-represented in polls in the 50s, Trump supporters are under-represented in the polls in the 20s, it's not the pollster's fault, they can be do everything near perfectly and this will still happen.

If I had to guess the polls are off by at least 10 points this time, if Joe is +2 he is actually -8, etc.