Matthew Barnett answers Is AI safety research less parallelizable than AI research?