Bias detection and mitigation: Practice Questions — Safety, Ethics and Compliance (NVIDIA-Certified Professional: Generative AI LLMs)

Practice Questions: Bias Detection and Mitigation for NVIDIA-Certified Professional: Generative AI LLMs

This set of multiple-choice questions is designed to help candidates prepare for the Safety, Ethics and Compliance section of the NVIDIA-Certified Professional: Generative AI LLMs exam, focusing specifically on bias detection and mitigation.

1. Which of the following best describes bias auditing in the context of large language models?

   • A. Training the model on a larger dataset to improve accuracy
   • B. Systematically evaluating the model outputs to identify unfair or prejudiced behavior
   • C. Increasing model size to reduce bias
   • D. Using guardrails to restrict model responses

   Correct answer: B

   Explanation: Bias auditing involves systematically evaluating model outputs to detect unfair or prejudiced behavior, which is essential for assessing fairness in LLMs.

2. Which technique is most effective for mitigating bias after detecting it in a trained LLM?

   • A. Increasing the number of training epochs
   • B. Applying post-processing adjustments to model outputs
   • C. Reducing model parameters
   • D. Ignoring biased outputs as rare cases

   Correct answer: B

   Explanation: Post-processing adjustments such as output filtering or re-ranking can mitigate bias by modifying outputs without retraining the entire model.

3. What is a common quantitative metric used in bias detection for language models?

   • A. BLEU score
   • B. Word error rate
   • C. Demographic parity difference
   • D. Perplexity

   Correct answer: C

   Explanation: Demographic parity difference measures fairness by comparing model outcomes across demographic groups, making it a key metric for bias detection.

4. Which of the following is an example of a guardrail implementation to prevent biased outputs?

   • A. Increasing dataset size
   • B. Adding explicit content filters to block harmful stereotypes
   • C. Using larger batch sizes during training
   • D. Reducing model depth

   Correct answer: B

   Explanation: Guardrails like content filters help prevent biased or harmful outputs by restricting certain types of responses.

5. During bias detection, why is it important to use diverse test datasets?

   • A. To improve model training speed
   • B. To ensure the model performs well only on common cases
   • C. To reveal biases that affect different demographic groups
   • D. To reduce the model size

   Correct answer: C

   Explanation: Diverse test datasets expose the model to varied demographic contexts, helping identify biases that may not appear in homogeneous data.

6. Which approach helps reduce bias during the training phase of a generative AI LLM?

   • A. Using biased datasets intentionally
   • B. Data augmentation with balanced demographic representation
   • C. Avoiding any data preprocessing
   • D. Training on only one demographic group

   Correct answer: B

   Explanation: Data augmentation with balanced representation helps the model learn fair patterns and reduces bias introduced by unbalanced training data.

7. What is the primary goal of bias mitigation in generative AI LLMs?

   • A. To eliminate all errors in model output
   • B. To ensure model outputs are fair and do not perpetuate harmful stereotypes
   • C. To increase model complexity
   • D. To reduce training time

   Correct answer: B

   Explanation: Bias mitigation aims to produce fair outputs and avoid reinforcing harmful stereotypes or unfair treatment of any group.

8. Which of the following is a limitation of automated bias detection tools?

   • A. They can perfectly identify all bias types
   • B. They may miss subtle or context-dependent biases
   • C. They eliminate the need for human review
   • D. They always improve model accuracy

   Correct answer: B

   Explanation: Automated tools may fail to detect subtle or context-specific biases, so human oversight remains critical.

More in this topic

Safety, Ethics and Compliance — NVIDIA-Certified Professional: Generative AI LLMsBias detection and mitigation — Safety, Ethics and Compliance (NVIDIA-Certified Professional: Generative AI LLMs)Bias detection and mitigation: Worked Example — Safety, Ethics and Compliance (NVIDIA-Certified Professional: Generative AI LLMs)Bias auditing and fairness assessment — Safety, Ethics and Compliance (NVIDIA-Certified Professional: Generative AI LLMs)Guardrails implementation — Safety, Ethics and Compliance (NVIDIA-Certified Professional: Generative AI LLMs)Bias detection and mitigation: Quick Reference — Safety, Ethics and Compliance (NVIDIA-Certified Professional: Generative AI LLMs)Bias detection and mitigation: Common Mistakes — Safety, Ethics and Compliance (NVIDIA-Certified Professional: Generative AI LLMs)