OpenAI researchers are developing a method to predict how often a new AI model will make mistakes before launch.

This could fill gaps left by standard safety testing.

AI-generated: summaries written by AI from the linked sources. How we use AI