Working through the formal foundations of provably beneficial agents - assistance games, provable guarantees of alignment and modern deep learning primitives.