Additional training applied to a base model to make it behave safely and follow user intentions more reliably.
Adhering to complex, structured, or constrained instructions