> It's like how software engineers who are at Google are worse if they're better competitive programmers.

That's not true. It didn't replicate and Norvig has said as much somewhere on HN, IIRC.

(I also agree with the other criticisms that this 'old vs young' setup in OP is obviously at least partially, and perhaps entirely, regression to the mean and Berkson.)

Makes sense. My perspective is that fast learners are fast because they absorb information quickly without the overhead of cross-domain synthesis. They have more logical contradictions in their minds which they haven't resolved or aren't even aware of. Their worldview is not coherent as a whole. In some cases, they don't have a worldview; instead they just rely on expert data to inform their decisions... But the experts themselves are often victim to the same kind of domain-specific tunnel vision. Such people often lack creativity in their work because cross-domain pattern synthesis is precisely how you can solve complex problems that haven't been solved before.