Your description of AlphaStar is wrong; that’s the architecture for AlphaGo. One network to evaluate positions and one to suggest good moves. IIRC AlphaStar has three networks which are (roughly) build, move, look.
Current theme: default
Less Wrong (text)
Less Wrong (link)
Your description of AlphaStar is wrong; that’s the architecture for AlphaGo. One network to evaluate positions and one to suggest good moves. IIRC AlphaStar has three networks which are (roughly) build, move, look.