RT @demishassabis: #DeepMind safety paper in collab w/@FHIOxford on making agents that can be safely interrupted: https://t.co/Q91Nt8Ykxq -…