“The current research direction of Paul Christiano and the Alignment Research Center, focusing on 'heuristic arguments' and mathematical proofs, is unlikely to solve alignment because deep learning progress has historically been empirical.”