Oh come on, guys. Sure, sampling from the tails of the distribution means you sample points with low probability, but if you sample them with disproportionatel

Oh come on, guys. Sure, sampling from the tails of the distribution means you sample points with low probability, but if you sample them with disproportionately low probability, you're simply failing to accurately sample the distribution. Translated: if you don't try crazy ideas sometimes, you're not accurately modeling the chances an idea is crazy.