The practice of structuring system instructions and delimiters (such as XML tags) to prevent users from hijacking or overriding model behaviors.
Want to actually apply concepts like this instead of just reading definitions?