Safety Benchmarking Quiz

What you'll be asked

1. Which of the following is the primary purpose of a 'Red Teaming' exercise in GenAI safety benchmarking?
2. When evaluating a model for 'jailbreak' susceptibility, which technique involves providing the model with a complex, multi-step scenario designed to bypass safety filters?
3. What is the primary advantage of using an 'LLM-as-a-Judge' approach for safety evaluation compared to human-only evaluation?
4. In the context of safety benchmarking, what does the 'Jailbreak Success Rate' (JSR) metric actually quantify?
5. When designing an automated safety evaluation pipeline, why is it critical to use a 'Constitutional AI' approach when selecting your judge model?

Answers and explanations come after you submit.