OpenAI is developing a superalignment program to control superintelligent AI systems and align them with human goals. The team is using a weak AI model to supervise a stronger one, with the aim of the stronger model learning from the weaker one’s mistakes. The first experiment showed promising results, particularly in natural language processing tasks. OpenAI is also offering $10 million in grants for work on various alignment approaches, encouraging the wider research community to contribute to this ambitious project.
OpenAI is developing a superalignment program to control superintelligent AI systems and align them with human goals. The team is using a weak AI model to supervise a stronger one, with the aim of the stronger model learning from the weaker one’s mistakes. The first experiment showed promising results, particularly in natural language processing tasks. OpenAI is also offering $10 million in grants for work on various alignment approaches, encouraging the wider research community to contribute to this ambitious project.