This is an early step toward more robustly beneficial and aligned models: training models to carry beneficial traits into new situations, so as AI becomes more capable, it also becomes more reliable, transparent, and helpful for people.

→ View original post on X — @openai