Learning more about how an LLM's output can be manipulated, because one is interested in executing such manipulation and/or because one is interested in preventing such manipulation.