[jamilton]

Why would that be different from generating 10 results and picking the top 2? I assume any 2 generated results from the same generation would be "within the same distribution", if that makes sense.

Is there a term for that approach?

[AlotOfReading]

Larger population = faster state space exploration. Sampling relies on the structure of the population. Most individuals will typically be minor variations on one of the successful genotypes, with some smaller percentage of novel genotypes. Random sampling will usually ensure that the successful genes are broadly represented while occasionally including the others. Since the sampling probably won't include the most optimized individuals, those others can compete in an easier environment and the algorithm has a better chance to escape local minima.