The 538 predictions are essentially the same as my playoff predictions. They are the best empirical estimate of the odds of the outcome of a race at the time the estimate is made. They include a degree of uncertainty, hence why the favorites are given odds of winning rather than saying candidate or team x is going to win for sure.
To anyone pointing at Romney's resurgence as evidence of the fallibility of the system, doesn't understand what the method claims to do. A real life event changed the odds, namely, Romney trouncing Obama in the first debate. No unbiased observers were predicting that to happen. The magnitude of that event was akin to a huge upset in the NFL. If the two had battled to a standoff, the race would not have tightened and Obama would probably have easily won. So back when Obama had an 80+% chance of winning the election, it meant that in 5 similar presidential races, the trailing candidate would typically win about 1 in 5 and that some game changing event would need to happen.
So think of these prediction tools as an instantaneous snapshot of the current situation. Understand that the odds are just that, and that the underdog sometimes wins. If you view them with that kind of understanding, these tools are pretty dang useful.