Known for 'Constitutional AI' training, which hardcodes safety principles directly into the model's objective function to reduce deceptive behavior.
Open the full topic