The AI Superalignment Problem

And the problem with the AI superalignment problem

Last week, there were a number of doomsday headlines around the departure of some high-level members of OpenAI who worked on "superalignment." Aligning AI with human values has long been a important part of AI research, often termed AI safety, ethical AI, or simply useful AI. AI systems can pose significant risks, so making sure that they are safe for h…