Our approach to alignment research

In three linesOpenAI improves its AI systems' ability to learn from human feedback and assist humans in evaluating AI. Goal: build a sufficiently aligned AI system to solve all other alignment problems.

Summary generated by Claude — human-verified