Polls don't purport to predict elections; though people do poll-based forecasts.
538, for instance, does a lot of them every cycle. And seems to be fairly close on the odds, though generally the most-likely predicted outcome occurs at a slightly higher rate than their predictions.
Notably, polls were generally fairly accurate for the US 2018 midterms.
In general, polls work better in electoral systems where national polls are useful; the US presidential election in particular is challenging.
There was a narrative on e-day that Trafalgar had done a good job, because Democratic votes were disproportionately counted later in the week. But if you'd relied on Trafalgar predictions to allocate resources in an election, you'd have gotten depantsed.
https://www.mediaite.com/tv/trafalgar-pollster-robert-cahaly...
The national polling firms are only getting airplay if they give the "right" result, the few groups that have been pretty accurate in 2016, 2018, and 2020 got very little media coverage due to not being close to the high profile ones and not having the results that CNN, ABC, NBC, etc. wanted to broadcast.
That coupled with comments from some prominent Democratic politicians that they want to make lists of Trump supporters presumably to punish them for their actions, has a significant chilling effect on people being willing to vocalize their support for Trump to non-friendly or unknown people.
Polling is starting to feel more like a weather forecast - presented with absolute certainty, only right about half the time.