Noticing confusion is the first skill I tried to train up last year, and is definitely a big one, because knowing what your models predict and noticing when they fail is a very valuable feedback loop that prevents you from learning if you can't even notice it.

Picturing what sort of evidence would unconvince you of something you actively believe is a good exercise to pair with the exercise of picturing what sort of evidence would convince you of something that seems super unlikely. Noticing unfairness there is a big one.

Realizing when you are trying to "win" at truthfinding, which is... ugh.