Paul Graham is not some magic oracle. He's a statistician who has seen, experienced and analyzed (formally and informally) a lot of very rich data. To ignore him and get away with it you either have to be an outlier in his model or the kind of person who would never appear in his model to begin with.

He mainly sees YC applicants and there may be a selection bias among them. Entirely possible that there is a really great trait among founders that makes them unlikely to apply to YC and likely to succeed without his advice.

That said, dude has been around.