How can a seemingly very intelligent man have such a poor understanding of probability

[TwilightZone]

7. The projections were wrong in part because of faulty assumptions.

538, for example, has noted that their models factored in the possibility that Biden would withdraw prior to SC, which was never going to happen. That caused the overall model and all future state models to be skewed.

"Ever since he finished fourth in Iowa, Biden has no longer been the favorite in South Carolina, according to our model. Sanders currently has a 1 in 2 (47 percent) chance of winning South Carolina, while Biden has a 2 in 5 (37 percent) shot. However, part of the reason our model has Sanders as the favorite is that it thinks Biden could drop out before South Carolina even votes. In the scenarios where Biden is still in the race come Feb. 29, though, he is probably still favored in the Palmetto State."

https://fivethirtyeight.com/features/election-update-biden-hasnt-exactly-collapsed-since-new-hampshire/

We have a much clearly picture of the race now. Biden is projected to win nearly every remaining state, and there's little reason to believe that they're as wrong as the SC projections.