Nate Silver blogs about the emerging split between robopolls and live-interviewer polls; the former lean much more heavily towards the Republicans than the latter. This complicates the process of deciding how to weight the polls when building election models.
Believe it or not, there's no particular reason to think that live interviewer polls are more accurate. Intuitively, it seems like they should be, and indeed, thinks like online polls are usually next to useless because the people who bothered to answer them are not a representative sample of the population. But neither are voters, exactly--and robopolls may actually select out the motivated voters who are most likely to actually make it to the polls.
We'll know on Tuesday night, one way or another. But this is going to have implications beyond politics, as politicians are not the only people who want to find out what others think of them. Changes in American lifestyles, from telecommuting to cell phones to civic engagement, are challenging traditional survey methods. This is going to affect social science, business, and of course, politicians and the journalists who love to follow them.