100% Latest & Updated Google Generative AI Leader Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 570
A new foundation-model release breaks a structured JSON output requirement. Which QA step should flag this?
- Evaluation after updates
- Performance monitoring
- Model version registry
- Security patch management
Explanation
Correct Answer
A:Regression evaluation tests established output contracts. Post-update evaluation checks regressions and changed behavior after a model or application update.
Incorrect Answers
B:Performance monitoring tracks task quality, response latency and service reliability in deployment. This does not directly meet the case requirement to verify formatting after a model update.
C:A registry records model versions and deployment metadata for reproducibility. This does not directly meet the case requirement to verify formatting after a model update.
D:Patch management applies supported fixes to model-serving components and dependencies. This does not directly meet the case requirement to verify formatting after a model update.
Question 571
An HR policy bot must answer using frequently updated source documents. What technique should be prioritized?
- Model evaluation
- Grounding through RAG
- Prompt engineering
- Fine-tuning
Explanation
Correct Answer
B:RAG supplies current internal text without repeated retraining. RAG retrieves current authorized passages at response time to support factual answers.
Incorrect Answers
A:Evaluation measures whether the proposed mitigation improves task performance. This does not directly meet the case requirement to answer from changing approved policies.
C:Prompt design improves task instructions and formatting without updating model parameters. This does not directly meet the case requirement to answer from changing approved policies.
D:Fine-tuning adapts model behavior through targeted training-weight updates. This does not directly meet the case requirement to answer from changing approved policies.
Question 572
An assistant keeps producing paragraphs when the user needs a three-field JSON object. Which first adjustment is reasonable?
- Human-in-the-loop review
- Prompt engineering
- Grounding through RAG
- Safety filtering
Explanation
Correct Answer
B:Explicit prompt constraints can steer output format. Prompt design improves task instructions and formatting without updating model parameters.
Incorrect Answers
A:Human review adds judgment and approval to sensitive decisions. This does not directly meet the case requirement to obtain a specified response structure.
C:RAG retrieves current authorized passages at response time to support factual answers. This does not directly meet the case requirement to obtain a specified response structure.
D:Safety filters screen for selected harmful content categories. This does not directly meet the case requirement to obtain a specified response structure.
Question 573
A model fails on a specialized recurring classification style despite adequate instructions and curated training cases. Which approach can adapt it?
- Grounding through RAG
- Prompt engineering
- Human-in-the-loop review
- Fine-tuning
Explanation
Correct Answer
D:Fine-tuning changes learned task behavior. Fine-tuning adapts model behavior through targeted training-weight updates.
Incorrect Answers
A:RAG retrieves current authorized passages at response time to support factual answers. This does not directly meet the case requirement to teach a stable domain-specific pattern.
B:Prompt design improves task instructions and formatting without updating model parameters. This does not directly meet the case requirement to teach a stable domain-specific pattern.
C:Human review adds judgment and approval to sensitive decisions. This does not directly meet the case requirement to teach a stable domain-specific pattern.
Question 574
A model proposes high-impact medical decisions. Which oversight mechanism should be required?
- Grounding through RAG
- Human-in-the-loop review
- Prompt engineering
- Safety filtering
Explanation
Correct Answer
B:Qualified review provides an accountable safety gate. Human review adds judgment and approval to sensitive decisions.
Incorrect Answers
A:RAG retrieves current authorized passages at response time to support factual answers. This does not directly meet the case requirement to have a professional assess sensitive output.
C:Prompt design improves task instructions and formatting without updating model parameters. This does not directly meet the case requirement to have a professional assess sensitive output.
D:Safety filters screen for selected harmful content categories. This does not directly meet the case requirement to have a professional assess sensitive output.
Question 575
A public chatbot emits abusive responses to provocative prompts. Which technical layer can help screen them?
- Grounding through RAG
- Human-in-the-loop review
- Safety filtering
- Model evaluation
Explanation
Correct Answer
C:Safety controls can block categories of unacceptable content. Safety filters screen for selected harmful content categories.
Incorrect Answers
A:RAG retrieves current authorized passages at response time to support factual answers. This does not directly meet the case requirement to reduce harmful output in user-facing responses.
B:Human review adds judgment and approval to sensitive decisions. This does not directly meet the case requirement to reduce harmful output in user-facing responses.
D:Evaluation measures whether the proposed mitigation improves task performance. This does not directly meet the case requirement to reduce harmful output in user-facing responses.
Question 576
A team adds RAG and wants proof that wrong citations decrease. Which discipline checks effectiveness?
- Safety filtering
- Fine-tuning
- Model evaluation
- Prompt engineering
Explanation
Correct Answer
C:Evaluation compares before-and-after grounded accuracy. Evaluation measures whether the proposed mitigation improves task performance.
Incorrect Answers
A:Safety filters screen for selected harmful content categories. This does not directly meet the case requirement to measure whether mitigation solved the problem.
B:Fine-tuning adapts model behavior through targeted training-weight updates. This does not directly meet the case requirement to measure whether mitigation solved the problem.
D:Prompt design improves task instructions and formatting without updating model parameters. This does not directly meet the case requirement to measure whether mitigation solved the problem.
Question 577
A financial agent confuses current policy rates because its training data is old. Which approach helps?
- Prompt engineering
- Grounding through RAG
- Human-in-the-loop review
- Model evaluation
Explanation
Correct Answer
B:RAG connects responses to current verified rates. RAG retrieves current authorized passages at response time to support factual answers.
Incorrect Answers
A:Prompt design improves task instructions and formatting without updating model parameters. This does not directly meet the case requirement to ground answers in fresh factual evidence.
C:Human review adds judgment and approval to sensitive decisions. This does not directly meet the case requirement to ground answers in fresh factual evidence.
D:Evaluation measures whether the proposed mitigation improves task performance. This does not directly meet the case requirement to ground answers in fresh factual evidence.
Question 578
A marketing model uses the wrong tone until the user clarifies role and audience. Which technique should be tried?
- Fine-tuning
- Grounding through RAG
- Human-in-the-loop review
- Prompt engineering
Explanation
Correct Answer
D:Prompt instructions can express audience and voice. Prompt design improves task instructions and formatting without updating model parameters.
Incorrect Answers
A:Fine-tuning adapts model behavior through targeted training-weight updates. This does not directly meet the case requirement to steer tone without model retraining.
B:RAG retrieves current authorized passages at response time to support factual answers. This does not directly meet the case requirement to steer tone without model retraining.
C:Human review adds judgment and approval to sensitive decisions. This does not directly meet the case requirement to steer tone without model retraining.
Question 579
A company has many validated examples of a specialist writing style that prompts cannot reliably reproduce. Which method might help?
- Model evaluation
- Human-in-the-loop review
- Grounding through RAG
- Fine-tuning
Explanation
Correct Answer
D:Fine-tuning adapts a model on those examples. Fine-tuning adapts model behavior through targeted training-weight updates.
Incorrect Answers
A:Evaluation measures whether the proposed mitigation improves task performance. This does not directly meet the case requirement to make specialized behavior persist across requests.
B:Human review adds judgment and approval to sensitive decisions. This does not directly meet the case requirement to make specialized behavior persist across requests.
C:RAG retrieves current authorized passages at response time to support factual answers. This does not directly meet the case requirement to make specialized behavior persist across requests.
Question 580
A hiring assistant suggests automatic rejection, but labor policy requires a recruiter to review. What safeguard applies?
- Prompt engineering
- Human-in-the-loop review
- Grounding through RAG
- Model evaluation
Explanation
Correct Answer
B:Human oversight preserves accountability for high-consequence outcomes. Human review adds judgment and approval to sensitive decisions.
Incorrect Answers
A:Prompt design improves task instructions and formatting without updating model parameters. This does not directly meet the case requirement to route employment decisions to human reviewers.
C:RAG retrieves current authorized passages at response time to support factual answers. This does not directly meet the case requirement to route employment decisions to human reviewers.
D:Evaluation measures whether the proposed mitigation improves task performance. This does not directly meet the case requirement to route employment decisions to human reviewers.
Generative AI Leader Practice Tests
