I don't like when "don't trust the polls because Brexit/Trump" gets trotted out, because neither of those was actually a polling upset. Brexit was a coin toss in the polls by the last few weeks, and we got a coin toss result. And the polls got Clinton v Trump also generally pretty accurate - the only reason he won was because of extremely thin margins in three states, something like 70k total, which was always how he was most likely to win if he was going to win. The reason those were upsets wasn't because of polling, but because Remain and Clinton were the consensus winners among most of the mainstream media and political classes of both the UK and US, and of most people in those countries, but that's not the same as the polls being wildly wrong.
At the same time, I think a huge amount of blame for why there was so much shock at Trump specifically winning is that the US media's way of interpreting and graphically representing statistical likelihood from polling pre-2016 was complete dogshit. You had outlets which should have known better putting out completely irresponsible lines like "Clinton has a 98% chance of winning", or using infographics that made a Clinton win seem inevitable. And we're talking about the US as well, where presidential elections that aren't close are historically rare, as you'd expect for a country with such a rigid two-candidate system. So now you have this folk memory that "the polls got it wrong", when really what happened was part of the larger systemic issue in the US media when it came to reporting on that election. This year, if anything, much of the media is now hedging its bets too much. Short of Biden lighting a molotov cocktail with an American flag as a fuse and throwing it into a police station, he's got this in terms of votes. A popular vote margin of around 4-5% or higher is more than enough to overcome any electoral college disadvantage, and even within the standard margin of error, at the worst, on current numbers he'd come above that. The problem is voter suppression, disqualification of mail-in ballots, and other ways the GOP is trying to rig the election. That's happening on a scale not seen since Jim Crow, and the modern US media - especially polling analysts - isn't really able to model it.
Actual polling mistakes in recent years are the 2015 and 2017 elections in the UK (although you could make the argument that the latter wasn't a miss as much as an exceptionally late surge that polling couldn't possibly have picked up in time). And part of the reason the Brexit polls were generally quite good in 2016 is because the British polling industry collectively revised much of their work after 2015. And 2019 then turned out to also be pretty much bang on in terms of polling, but it only felt surprising because of the media and popular perception that polling has just completely stopped being trustworthy.
