Reminds me of how the AlphaGo imstance that trained against human experts was equalled or bested by the version that just bootstrapped by playing against itself

Reminds me of how the AlphaGo imstance that trained against human experts was equalled or bested by the version that just bootstrapped by playing against itself. (With the math side being analagous to the self-trained instance)