Balance and de-bias the training corpus to reduce harmful skew and Add fairness metrics to the model evaluation process are correct because responsible AI for policy review must address data harms and measure outcomes to confirm fairness. Teams can use tools such as Amazon SageMaker Clarify to detect and monitor bias during data preparation and model evaluation.

Balance and de-bias the training corpus to reduce harmful skew reduces systematic errors that could disadvantage particular groups of policyholders and it improves the validity of downstream fairness checks. Add fairness metrics to the model evaluation process creates measurable signals that governance teams can audit and that guide mitigation efforts before deployment.

Tune the temperature parameter to 0.2 for more consistent responses is about generation variability and it does not change biased patterns in the training data nor provide accountability for fairness.

Use prompt engineering to coax the model toward desired formats can help shape outputs but it is not a substitute for fixing biased data or for measuring fairness during evaluation.

Train longer to memorize domain terminology and reduce loss risks overfitting and it does not inherently address fairness or ethical risks so it is not a responsible mitigation strategy on its own.

Full AWS Practitioner Certification Question