[dontreact]

They can always finetune using RL later. Superversied training was the first step at making AlphaGo work.