Describe AI MLOps/LLMOps Ecosystems and General Use CERTIFIED-AI-EXPERT Practice Questions (Page 3)

Questions 11–15

1. 11expert · hard

   A retail bank is adopting MLOps for its loan approval models. The bank must ensure that the models do not discriminate against protected groups. The data science team is considering various approaches to monitor for bias. Which approach is most effective?

   • ARelying on the model's training data being unbiased, assuming the model will not learn biases.
   • BRegularly evaluating model predictions across different demographic groups and comparing approval rates.
   • CUsing a more complex model that is less likely to exhibit bias.
   • DMonitoring only the overall accuracy of the model, without breaking down by group.
   Reveal Solution Discussion Select an answer first
2. 12application · medium

   A marketing agency wants to use LLMOps to generate personalized ad copy for different customer segments. They need to test multiple prompt variations and measure which ones drive the highest click-through rates. Which practice is most aligned with this use case?

   • ADeploying the model without any tracking, relying on overall sales figures to judge success.
   • BManually reviewing the generated ad copy and selecting the best one based on intuition.
   • CImplementing an A/B testing framework that tracks prompt versions and their performance metrics.
   • DUsing a single, fixed prompt for all customer segments to ensure consistency.
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   Which lifecycle stage in MLOps involves tracking model performance and detecting data drift?

   • AData preparation
   • BModel monitoring
   • CFeature engineering
   • DModel deployment
   Reveal Solution Discussion Select an answer first
5. 14expert · hard

   A logistics company is implementing MLOps for its route optimization models. The models are retrained daily with new traffic and delivery data. The team is debating between using a fully automated pipeline and a human-in-the-loop approach. Which consideration is most important when deciding between the two?

   • AThe speed of the training pipeline, as faster is always better.
   • BThe cost of the cloud infrastructure, which should be minimized at all costs.
   • CThe preference of the data science team for either full automation or manual review.
   • DThe level of risk associated with model errors and the need for human oversight.
   Reveal Solution Discussion Select an answer first
6. 15application · medium

   A media company is deploying a recommendation system that uses both collaborative filtering and a large language model to generate personalized summaries. They want to monitor the system's performance in production and detect when the model's recommendations become stale or biased. Which monitoring approach is most appropriate?

   • AReview the model's outputs manually on a quarterly basis.
   • BMonitor only the infrastructure metrics, such as CPU and memory usage, to ensure the system is running.
   • CTrack both model performance metrics (e.g., click-through rate) and data drift indicators, with alerts for significant deviations.
   • DOnly monitor the LLM's response length and latency, assuming that if these are normal, the model is working well.
   Reveal Solution Discussion Select an answer first