Training techniques used to make a model refuse harmful requests and behave responsibly, reducing the risk of misuse.
Adhering to complex, structured, or constrained instructions