Lots of RV polls (which show Biden with a big lead) are weighted based on census data. Which means they show what the results of the election would look like IF the electorate looked demographically exactly like the census...

If you weighted the same poll using the 2016 electorate, you would see Biden's huge lead evaporate based on nothing except how you chose to manipulate the poll...

Read this awesome article. It's ONE set of 2016 polling data weighted by 5 different pollsters. The results range from Clinton +4 to Trump +1, and they're ALL using the exact same data: https://www.nytimes.com/interactive/2016/09/20/upshot/the-error-the-polling-world-rarely-talks-about.html

In order to get the polling right, you have to be able to know what the electorate will look like and literally NOBODY knows what the electorate will look like.