A generative assistant is heading toward a production rollout. Which task should be included in its pre-launch evaluation plan?
Select an answer to reveal the explanation.
Short Explanation
You need to test what the assistant actually says before rollout. Run a representative set of prompts and check groundedness, relevance, harmful content, and policy compliance. Capacity and naming don't prove the answers are safe.
Full Explanation
A pre-launch evaluation plan should test what the system actually produces against the criteria that determine whether an answer is trustworthy and safe to show a user. The task to include is testing a representative set of prompts for groundedness, relevance, harmful content, and policy compliance. Confirming that the model's parameter count meets a minimum threshold is incorrect because parameter count is a technical property with no direct bearing on output quality or safety. Confirming that the deployment infrastructure has sufficient compute capacity is incorrect because infrastructure capacity is an operational readiness concern, separate from content evaluation. Confirming that the assistant's name has been trademarked is incorrect because trademark status has nothing to do with whether the assistant's outputs are grounded or safe. For the exam, build the evaluation plan around content criteria: groundedness, relevance, harm, and compliance.