3.1 Describe How to Proactively Overcome Foundation Model Limitations. GENERATIVE-AI-LEADER Practice Questions (Page 1)

Questions 1–5

1. 1application · medium

   A news organization is using a foundation model to generate article summaries. The model frequently produces summaries that sound confident but contain factual errors about recent events. The organization needs to improve accuracy without waiting for a new model release. What is the most effective immediate mitigation?

   • ASwitch to a smaller model that has a more recent training cutoff.
   • BGround the model with a real-time news database using RAG before generating summaries.
   • CUse a lower temperature to make the model more deterministic and factual.
   • DIncrease the model's max output tokens to allow for more detailed summaries.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A university is deploying a generative AI tutor to help students with physics homework. The tutor sometimes gives incorrect answers on advanced topics. The team has a small set of verified example problems and solutions. They want to improve accuracy without a large fine-tuning effort. Which approach is most appropriate?

   • AIncrease the model's temperature to encourage more creative problem-solving.
   • BUse few-shot prompting with the verified examples to guide the model's reasoning.
   • CUse a different foundation model that is larger and more expensive.
   • DFine-tune the model on the small set of examples to specialize it for physics.
   Reveal Solution Discussion Select an answer first
4. 3expert · hard

   A government agency is deploying a generative AI system to summarize public comments on proposed regulations. The system must be fair and unbiased across different demographic groups. The team has a monitoring dashboard but is unsure which metrics to track. Which combination of practices best addresses the bias and fairness limitation?

   • ARestrict the model's input to only the most common comment topics to reduce complexity.
   • BFine-tune the model on all public comments equally to ensure balanced representation.
   • CUse a higher temperature to generate more diverse summaries across all groups.
   • DEvaluate model outputs across demographic subgroups and track performance metrics for each group over time.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   A team has deployed a generative AI model and wants to detect when the model's performance degrades over time due to changes in the input data distribution. Which practice is most directly relevant?

   • ASecurity patches
   • BAutomatic model upgrades
   • CVersioning
   • DDrift monitoring
   Reveal Solution Discussion Select an answer first
6. 5foundation · easy

   A generative AI model trained on data up to 2022 is asked about a product released in 2024. Which foundation model limitation does this scenario best illustrate?

   • ABias
   • BKnowledge cutoff
   • CData dependency
   • DHallucination
   Reveal Solution Discussion Select an answer first