“While AlphaGo Lee used two separate neural networks for policy and value, subsequent versions merged them into a single network with two output heads.”