Irving Sees Hope in Machine-Aided Alignment Research
UK AISI chief scientist Geoffrey Irving outlines one path for alignment research on short timelines.
TLDR
Geoffrey Irving, Chief Scientist at the UK AI Security Institute, replied that recent models handle both prose and formal mathematics well. He described a third hope for conceptual alignment research: humans assign as many subtasks as possible to machines. Irving stated this approach could stay meaningful even under short timelines, drawing on his earlier roles at DeepMind, OpenAI, and Google Brain.
Combined views
548
1 Source, first seen 26d ago