A proposed method for keeping AI systems safe in lack of unified efforts.

We gain a lead in capabilities, and use that lead to:

  • Automate alignment research
  • Prevent others from building unsafe systems