[cpp_frog]

Anything that isn't scalable. Taleb writes at length about this difference in The Black Swan, categorizing professions in which there are winner-take-all effects (Extremistan) and low variance, stable, one-sample-is-representative ones (Mediocristan). As you said, surgeon (the example Taleb gives is 'dentist'), prostitute, taylor, and so on. He also notes that some professions can switch categories, e.g. musician, since the existence of recording technology the distribution became a Pareto one.