On RCP, their state specific polls have been PA Clinton +8 2 times. Their state specific GA poll was Trump +4.
Their "model" basically takes a huge national survey, use MRP (multilevel regression and post-stratification), taking various attributes of you, and then applying that to what they know about the states. So someone in Alaska could be affecting the Georgia results...
Basically I think they are using national white college support of Clinton (55%) too much in Georgia (where white college are only 30% for her in GA)...
https://today.yougov.com/news/2016/10/04/YouGov-Model-State-of-2016/
