large language models (LLMs) exhibit remarkable fluency and competence across
various natural language tasks. However, recent research has highlighted their
sensitivity to variations in input prompts. To deploy LLMs in a safe and
reliable manner, it is crucial for their outputs to be c