Back to the topic at hand, Silver has
this post about not counting Trump out too early. In it, he points out that going into Election Day, the 538 model also gave Trump a 29% chance of winning in 2016, but then goes on to describe how Trump's 29% in November 2016 is very different than Trump's 29% chance in August 2020.
Basically, Trump had a chance going into Election Day because the race was close, and a modest polling error in key states could easily swing him the election (which is, in fact, what happened). Now, however, Biden has a much bigger lead than Clinton did, and Trump's 29% chance of winning is rooted almost entirely in uncertainty. To put it simply, the model suspects there is a decent chance the race will become closer as the election approaches. But if it doesn't, if Biden maintains his current lead, then Biden's 71% odds of winning go up to something like 93%.
In other words, it's a good lesson that models are, to put it mildly, volatile. As the next few months go by, Biden could turn this into a rout, or Trump could close the gap. We still have the conventions and the debates and who knows what will happen with the economy and Covid. It's a model, not a prediction. That's a distinction that too many people didn't clearly make in 2016.