Amazon AWS Certified AI Practitioner AIF-C01 Practice Test 8
Question 70
A tax calculation uses a fixed statutory bracket and must always return the same exact result for the same numbers. A colleague suggests: Train a generative model to invent tax brackets from examples. Which finding BEST explains why the suggestion should be rejected?
- A generative prediction is unsuitable for fixed legal arithmetic and could hallucinate
- Image creation neither models transaction labels nor meets the required decision timing
- Polly generates speech from text rather than transcribing audio
- The shortcut is safe because its terminology is familiar to the team
Explanation
Correct Answer
A:A generative prediction is unsuitable for fixed legal arithmetic and could hallucinate. This is the direct defect in the proposed action: Train a generative model to invent tax brackets from examples.
Incorrect Answers
B:This finding is associated with a different problem: A finance firm has labeled historical credit-card transactions and wants real-time alerts on likely fraudulent purchases.
C:This concern belongs to another situation: A call center stores English phone recordings and needs searchable written transcripts of the speech.
D:The actual defect is: A generative prediction is unsuitable for fixed legal arithmetic and could hallucinate.
Question 71
A tax calculation uses a fixed statutory bracket and must always return the same exact result for the same numbers. After implementing Implement deterministic rules instead of probabilistic ML for this statutory calculation, which observation BEST verifies the intended result?
- Test inputs match the published statutory formula exactly
- Previously unseen test transactions receive measured fraud scores during authorization
- The team schedules a future test of whether test inputs match the published statutory formula exactly, but has no result
- The team approves Implement deterministic rules instead of probabilistic ML for this statutory calculation but does not run an operational test
Explanation
Correct Answer
A:The observed result, Test inputs match the published statutory formula exactly, provides direct evidence. Known precise arithmetic rules are more appropriate than uncertain predictions.
Incorrect Answers
B:This observation would be evidence for another case: A finance firm has labeled historical credit-card transactions and wants real-time alerts on likely fraudulent purchases, not the current scenario.
C:An unperformed future check is not evidence. Verify the actual outcome: Test inputs match the published statutory formula exactly.
D:Approval is a plan, not a demonstrated result. The required observable condition is: Test inputs match the published statutory formula exactly.
Question 72
The initial scenario was: A tax calculation uses a fixed statutory bracket and must always return the same exact result for the same numbers. The original control is already in place. Now the team next wants to predict which filings may need manual review. Which ADDITIONAL action BEST addresses the new condition?
- Evaluate an ML risk classification model with labeled review outcomes
- Test inputs match the published statutory formula exactly
- Implement deterministic rules instead of probabilistic ML for this statutory calculation
- Train a generative model to invent tax brackets from examples
Explanation
Correct Answer
A:The team next wants to predict which filings may need manual review. The appropriate added action is: Evaluate an ML risk classification model with labeled review outcomes.
Incorrect Answers
B:This describes evidence for the original approach, not a new action: Test inputs match the published statutory formula exactly.
C:This was the original control and is already implemented. Known precise arithmetic rules are more appropriate than uncertain predictions.
D:This original shortcut remains unsuitable: A generative prediction is unsuitable for fixed legal arithmetic and could hallucinate.
Question 73
A finance firm has labeled historical credit-card transactions and wants real-time alerts on likely fraudulent purchases. Which option BEST meets the stated AI/ML or AWS requirement?
- Use Amazon Transcribe for speech-to-text conversion
- Use batch image generation as the primary transaction fraud detector
- Use supervised fraud classification with fraud labels and a real-time decision path
- Add unsupervised anomaly or clustering analysis for exploratory patterns
Explanation
Correct Answer
C:Labeled event histories support predictive classification at authorization time. In this situation, choose: Use supervised fraud classification with fraud labels and a real-time decision path.
Incorrect Answers
A:This can be valid for a different task: A call center stores English phone recordings and needs searchable written transcripts of the speech. The current request is different and needs the selected approach.
B:Image creation neither models transaction labels nor meets the required decision timing. The proposal, Use batch image generation as the primary transaction fraud detector, does not satisfy this need.
D:This is appropriate only after a distinct change: Investigators later seek previously unknown transaction behavior clusters. It is not the best initial response to a finance firm has labeled historical credit-card transactions and wants real-time alerts on likely fraudulent purchases.
Question 74
A finance firm has labeled historical credit-card transactions and wants real-time alerts on likely fraudulent purchases. A colleague suggests: Use batch image generation as the primary transaction fraud detector. Which finding BEST explains why the suggestion should be rejected?
- Translate changes text languages and does not itself produce speech waveforms
- Image creation neither models transaction labels nor meets the required decision timing
- The shortcut is safe because its terminology is familiar to the team
- Polly generates speech from text rather than transcribing audio
Explanation
Correct Answer
B:Image creation neither models transaction labels nor meets the required decision timing. This is the direct defect in the proposed action: Use batch image generation as the primary transaction fraud detector.
Incorrect Answers
A:This concern belongs to another situation: An accessibility service has approved text instructions and must speak them aloud to callers.
C:The actual defect is: Image creation neither models transaction labels nor meets the required decision timing.
D:This finding is associated with a different problem: A call center stores English phone recordings and needs searchable written transcripts of the speech.
Question 75
A finance firm has labeled historical credit-card transactions and wants real-time alerts on likely fraudulent purchases. After implementing Use supervised fraud classification with fraud labels and a real-time decision path, which observation BEST verifies the intended result?
- The team schedules a future test of whether previously unseen test transactions receive measured fraud scores during authorization, but has no result
- Previously unseen test transactions receive measured fraud scores during authorization
- Searchable transcripts contain the recognized speech from sample recordings
- The team approves Use supervised fraud classification with fraud labels and a real-time decision path but does not run an operational test
Explanation
Correct Answer
B:The observed result, Previously unseen test transactions receive measured fraud scores during authorization, provides direct evidence. Labeled event histories support predictive classification at authorization time.
Incorrect Answers
A:Verify the actual outcome: Previously unseen test transactions receive measured fraud scores during authorization.
C:This observation would be evidence for another case: A call center stores English phone recordings and needs searchable written transcripts of the speech, not the current scenario.
D:The required observable condition is: Previously unseen test transactions receive measured fraud scores during authorization.
Question 76
The initial scenario was: A finance firm has labeled historical credit-card transactions and wants real-time alerts on likely fraudulent purchases. The original control is already in place. Now investigators later seek previously unknown transaction behavior clusters. Which ADDITIONAL action BEST addresses the new condition?
- Use supervised fraud classification with fraud labels and a real-time decision path
- Previously unseen test transactions receive measured fraud scores during authorization
- Use batch image generation as the primary transaction fraud detector
- Add unsupervised anomaly or clustering analysis for exploratory patterns
Explanation
Correct Answer
D:Investigators later seek previously unknown transaction behavior clusters. The appropriate added action is: Add unsupervised anomaly or clustering analysis for exploratory patterns.
Incorrect Answers
A:This was the original control and is already implemented. Labeled event histories support predictive classification at authorization time.
B:This describes evidence for the original approach, not a new action: Previously unseen test transactions receive measured fraud scores during authorization.
C:This original shortcut remains unsuitable: Image creation neither models transaction labels nor meets the required decision timing.
Question 77
An ML team evaluates a classifier only on records it used for model fitting and claims its score predicts future accuracy. Which option BEST meets the stated AI/ML or AWS requirement?
- Use the training score alone as the final production quality estimate
- Monitor postdeployment drift and retraining triggers
- Identify Amazon SageMaker AI as the managed ML development environment
- Reserve independent validation and test sets before assessing generalization
Explanation
Correct Answer
D:Unseen evaluation data exposes overfitting that training scores may hide. In this situation, choose: Reserve independent validation and test sets before assessing generalization.
Incorrect Answers
A:Training accuracy is optimistically biased and not independent evidence. The proposal, Use the training score alone as the final production quality estimate, does not satisfy this need.
B:This is appropriate only after a distinct change: A new model version is deployed and input patterns shift seasonally. It is not the best initial response to an ml team evaluates a classifier only on records it used for model fitting and claims its score predicts future accuracy.
C:This can be valid for a different task: A company needs managed tools for preparing, training, evaluating, and deploying custom predictive models on AWS. The current request is different and needs the selected approach.
Question 78
An ML team evaluates a classifier only on records it used for model fitting and claims its score predicts future accuracy. A colleague suggests: Use the training score alone as the final production quality estimate. Which finding BEST explains why the suggestion should be rejected?
- Text-to-speech service functionality is not a general ML training environment
- Training accuracy is optimistically biased and not independent evidence
- Changed data distributions can invalidate previous evaluation results
- The shortcut is safe because its terminology is familiar to the team
Explanation
Correct Answer
B:Training accuracy is optimistically biased and not independent evidence. This is the direct defect in the proposed action: Use the training score alone as the final production quality estimate.
Incorrect Answers
A:This finding is associated with a different problem: A company needs managed tools for preparing, training, evaluating, and deploying custom predictive models on AWS.
C:This concern belongs to another situation: A retailer deployed a model months ago and now sees prediction quality drop after customer habits change.
D:Familiar wording is not proof that an action is appropriate. The actual defect is: Training accuracy is optimistically biased and not independent evidence.
Question 79
An ML team evaluates a classifier only on records it used for model fitting and claims its score predicts future accuracy. After implementing Reserve independent validation and test sets before assessing generalization, which observation BEST verifies the intended result?
- The team demonstrates tracked experiments and a deployed predictive endpoint
- The final report includes untouched test-set performance and data-split lineage
- The team schedules a future test of whether the final report includes untouched test-set performance and data-split lineage, but has no result
- The team approves Reserve independent validation and test sets before assessing generalization but does not run an operational test
Explanation
Correct Answer
B:The observed result, The final report includes untouched test-set performance and data-split lineage, provides direct evidence. Unseen evaluation data exposes overfitting that training scores may hide.
Incorrect Answers
A:This observation would be evidence for another case: A company needs managed tools for preparing, training, evaluating, and deploying custom predictive models on AWS, not the current scenario.
C:Verify the actual outcome: The final report includes untouched test-set performance and data-split lineage.
D:The required observable condition is: The final report includes untouched test-set performance and data-split lineage.
Question 80
The initial scenario was: An ML team evaluates a classifier only on records it used for model fitting and claims its score predicts future accuracy. The original control is already in place. Now a new model version is deployed and input patterns shift seasonally. Which ADDITIONAL action BEST addresses the new condition?
- Monitor postdeployment drift and retraining triggers
- Reserve independent validation and test sets before assessing generalization
- The final report includes untouched test-set performance and data-split lineage
- Use the training score alone as the final production quality estimate
Explanation
Correct Answer
A:A new model version is deployed and input patterns shift seasonally. The appropriate added action is: Monitor postdeployment drift and retraining triggers.
Incorrect Answers
B:This was the original control and is already implemented. Unseen evaluation data exposes overfitting that training scores may hide.
C:This describes evidence for the original approach, not a new action: The final report includes untouched test-set performance and data-split lineage.
D:This original shortcut remains unsuitable: Training accuracy is optimistically biased and not independent evidence.