Sutskever co-led the superalignment team at OpenAI, which ensures that AI follows what humans want it to do even as it gets smarter.