OpenAIs Superalignment team, led by Ilya Sutskever, has developed a method to guide the behavior of increasingly intelligent AI models. The company aims to maintain control over AI systems that could become superhuman and potentially dangerous. Their research paper explores the use of supervision to tune AI systems and automate feedback processes as AI advances. Through experiments, they have found methods to prevent stronger AI models from becoming less capable when guided by weaker models. This research serves as a starting point for further investigations into controlling super-intelligent AI.
