Safety mechanisms built into a model that prevent it from generating harmful, inappropriate, or restricted content.
Adhering to complex, structured, or constrained instructions