Openai's Ilya Sutskever has a plan for keeping super- intelligent.

OpenAI was founded to build artificial intelligence that benefits all of humanity, even when it becomes considerably smarter than its creators. The company says that a new research group is starting to bear fruit. The Superalignment project will get a fifth of OpenAI's available computing power. The research paper released today by OpenAI shows the results of experiments that try to let an inferior model guide the behavior of a much smarter one. The scenario was designed to stand in for a future time when humans need to work with systems that are more intelligent than themselves. The process called supervision is used to tune systems like GPT-4 to be more helpful and less harmful. As the power of artificial intelligence increases, researchers think it may become impossible for humans to give useful feedback. The more recent system became less capable and similar to the inferior system during a control experiment. The team added an algorithmic tweak to GPT-4 that allowed the stronger model to follow the guidance of the weaker model without blunting its performance as much as would normally happen. The researchers admit that these methods do not guarantee that the stronger model will behave perfectly, but they say that this is a starting point for further research. The director of the Center for AI Safety in San Francisco says that it's great to see OpenAI address the problem of controlling artificial intelligence.