passdrill
AI & LLM Engineering · Evaluation, Testing & Safety · Card 005/010 easy

Before releasing a new model, a safety team assembles a group of people whose specific job is to deliberately try to provoke the model into producing harmful, disallowed, or policy-violating outputs, using creative and adversarial prompts, so the team can find and fix weaknesses before real users encounter them. What is this practice called, and how does it differ from routine functional testing?

  1. This is red-teaming; unlike routine functional testing, which checks whether the model performs its intended tasks correctly, red-teaming specifically and adversarially searches for inputs that cause unsafe or policy-violating behavior, deliberately trying to break the model's safety guardrails rather than verify normal functionality
  2. This is A/B testing; like routine functional testing, it measures average task success rate across two model versions shown to random subsets of real users, without any adversarial intent
  3. This is unit testing; it differs from routine functional testing only in that the test cases are written by an external team rather than the model's own developers
  4. This is regression testing; it differs from routine functional testing only in that it is run after every deployment rather than before the first one
Next card → Shuffle