AWS Certified AI Practitioner AIF-C01 Practice Test 40
Question 390
A GenAI assistant must avoid repeating social security numbers found in user-provided text. During a design review, someone proposes Increase model temperature so identifiers become random. Which technical objection is specifically relevant to this proposal?
- Sampling does not reliably prevent information disclosure
- A polite request is not enforceable protection
- Sensitive data handling needs explicit controls rather than model goodwill
- Unnecessary prompt logging can increase exposure
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Increase model temperature so identifiers become random. Sampling does not reliably prevent information disclosure
Incorrect Answers
B:This limitation concerns Rely only on a natural-language request to keep secrets; it does not diagnose the proposed Increase model temperature so identifiers become random approach.
C:Sensitive data handling needs explicit controls rather than model goodwill This explains the recommended design, but it is not the requested diagnosis of Increase model temperature so identifiers become random.
D:This limitation concerns Use public CloudTrail events as a repository for raw sensitive prompts; it does not diagnose the proposed Increase model temperature so identifiers become random approach.
Question 391
An analyst needs a model explanation that business users can understand without revealing training secrets. Which option BEST meets the business requirement?
- Reveal private training rows as the explanation
- Use only neural network layer counts for every decision
- Provide understandable feature attribution and a clear model decision summary
- Provide a billing receipt instead of decision rationale
Explanation
Correct Answer
C:Useful explainability describes factors affecting predictions in an appropriate form The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Exposing source records can violate privacy and still be confusing This prevents the approach from meeting the stated requirement.
B:Architecture size does not explain a particular prediction This prevents the approach from meeting the stated requirement.
D:Charges do not establish why the model responded This prevents the approach from meeting the stated requirement.
Question 392
A deployed predictor starts receiving different input distributions from the data used for validation. During a design review, someone proposes Retrain continuously without tracking data or performance changes. Which technical objection is specifically relevant to this proposal?
- Storage bytes do not diagnose feature distribution change
- Production input changes can degrade predictive behavior
- Deployment state says nothing about future distribution shift
- Uncontrolled retraining can introduce new risk
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Retrain continuously without tracking data or performance changes. Uncontrolled retraining can introduce new risk
Incorrect Answers
A:This limitation concerns Measure only S3 storage growth as model drift; it does not diagnose the proposed Retrain continuously without tracking data or performance changes approach.
B:Production input changes can degrade predictive behavior This explains the recommended design, but it is not the requested diagnosis of Retrain continuously without tracking data or performance changes.
C:This limitation concerns Assume a successful deployment permanently guarantees accuracy; it does not diagnose the proposed Retrain continuously without tracking data or performance changes approach.
Question 393
A regulated firm must document a model's intended use, limitations and risk rating. During a design review, someone proposes Assume a saved model weight file documents model lineage and risk. Which technical objection is specifically relevant to this proposal?
- Weights alone do not describe governance decisions
- Model Cards capture model metadata, intended uses and risk considerations
- Cost reports omit intended uses and technical limitations
- A storage path cannot capture evaluation and intended use
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Assume a saved model weight file documents model lineage and risk. Weights alone do not describe governance decisions
Incorrect Answers
B:Model Cards capture model metadata, intended uses and risk considerations This explains the recommended design, but it is not the requested diagnosis of Assume a saved model weight file documents model lineage and risk.
C:This limitation concerns Use billing invoices as the complete model-risk record; it does not diagnose the proposed Assume a saved model weight file documents model lineage and risk approach.
D:This limitation concerns Treat a dataset location as the entire model documentation; it does not diagnose the proposed Assume a saved model weight file documents model lineage and risk approach.
Question 394
A credit-screening model performs well overall but misses qualified applicants in one protected group. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- A system inventory lists this competing approach as active: Use overall accuracy as the only fairness measure
- A component test invokes this alternate approach successfully: Judge fairness solely by model parameter count
- A reviewed design record approves this competing approach: Remove demographic evaluation fields before measuring disparities
- Reviewed subgroup confusion matrices quantify the unequal false-negative rates
Explanation
Correct Answer
D:Reviewed subgroup confusion matrices quantify the unequal false-negative rates. This checks the business result rather than just a configured feature.
Incorrect Answers
A:A high average score can coexist with harmful group-level errors An active setting cannot independently prove the requested business outcome.
B:Size does not measure differential impact A component invocation does not validate the necessary end-to-end outcome.
C:Without appropriate controlled subgroup information the disparity may remain hidden Design approval is not evidence that the alternative meets the actual requirement.
Question 395
A language model is being adapted for a multilingual community with underrepresented dialects. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- A component test invokes this alternate approach successfully: Use only one high-performing standard-language benchmark
- A system inventory lists this competing approach as active: Train only on dominant dialect samples and assume transfer is perfect
- Native-dialect reviewers report acceptable output quality across sampled dialects
- A reviewed design record approves this competing approach: Increase output tokens as the entire inclusion strategy
Explanation
Correct Answer
C:Native-dialect reviewers report acceptable output quality across sampled dialects. This checks the business result rather than just a configured feature.
Incorrect Answers
A:A single dialect benchmark cannot prove broader coverage A component invocation does not validate the necessary end-to-end outcome.
B:Missing dialect examples can harm the omitted users An active setting cannot independently prove the requested business outcome.
D:More generated words cannot add missing training representation Design approval is not evidence that the alternative meets the actual requirement.
Question 396
A high-impact customer eligibility model has borderline cases needing a human decision. Which option BEST meets the business requirement?
- Use lower latency as proof that human review is unnecessary
- Treat a guardrail toxicity filter as the final eligibility reviewer
- Automatically approve every high-confidence output without defined controls
- Use a human-in-the-loop review workflow such as Amazon A2I where appropriate
Explanation
Correct Answer
D:Reviewers can inspect uncertain outputs before critical action The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Faster inference is not evidence of acceptable decision risk This prevents the approach from meeting the stated requirement.
B:Toxicity filtering does not replace human eligibility judgment This prevents the approach from meeting the stated requirement.
C:Confidence alone may not meet policy for sensitive outcomes This prevents the approach from meeting the stated requirement.
Question 397
A regulated firm must document a model's intended use, limitations and risk rating. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- A system inventory lists this competing approach as active: Use billing invoices as the complete model-risk record
- A component test invokes this alternate approach successfully: Assume a saved model weight file documents model lineage and risk
- Reviewers can inspect a current card containing purpose, metrics and risk
- A reviewed design record approves this competing approach: Treat a dataset location as the entire model documentation
Explanation
Correct Answer
C:Reviewers can inspect a current card containing purpose, metrics and risk. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Cost reports omit intended uses and technical limitations An active setting cannot independently prove the requested business outcome.
B:Weights alone do not describe governance decisions A component invocation does not validate the necessary end-to-end outcome.
D:A storage path cannot capture evaluation and intended use Design approval is not evidence that the alternative meets the actual requirement.
Question 398
An AI assistant's refusals frustrate legitimate users while missing some unsafe requests. Which option BEST meets the business requirement?
- Set all filters maximally restrictive without usability testing
- Tune guardrail policies using both false-positive and false-negative test cases
- Turn all safety filters off to eliminate refusals
- Judge only the total number of blocked prompts
Explanation
Correct Answer
B:Safety requires balancing blocked harmful output with useful allowed output The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Overblocking can prevent legitimate business activity This prevents the approach from meeting the stated requirement.
C:No filtering exposes users to prohibited content This prevents the approach from meeting the stated requirement.
D:Raw block count does not reveal missed harms or unjustified refusals This prevents the approach from meeting the stated requirement.
Question 399
A model offers high accuracy but opaque reasons for high-stakes decisions. Before approving a rollout, an architect suggests Use model popularity as the risk-adjusted decision. Which mismatch should be documented for this precise design?
- Interpretability alone does rather than guarantee acceptable outcomes
- Opaque gains may not justify lost explainability in regulated decisions
- Adoption counts do not measure explainability or model reliability
- Pure accuracy can fail explanation obligations
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Use model popularity as the risk-adjusted decision. Adoption counts do not measure explainability or model reliability
Incorrect Answers
A:This limitation concerns Choose the simplest model regardless of accuracy or safety; it does not diagnose the proposed Use model popularity as the risk-adjusted decision approach.
B:Opaque gains may not justify lost explainability in regulated decisions This explains the recommended design, but it is not the requested diagnosis of Use model popularity as the risk-adjusted decision.
D:This limitation concerns Choose highest accuracy without assessing auditability; it does not diagnose the proposed Use model popularity as the risk-adjusted decision approach.
Question 400
A customer-facing medical information assistant may invent unsafe dosage recommendations. Before approving a rollout, an architect suggests Increase completion length to improve medical authority. Which mismatch should be documented for this precise design?
- High-stakes unsupported medical claims need source checking and human control
- Removing provenance increases the chance of undetected errors
- Model confidence cannot establish medical safety
- Longer advice is not clinical validation
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Increase completion length to improve medical authority. Longer advice is not clinical validation
Incorrect Answers
A:High-stakes unsupported medical claims need source checking and human control This explains the recommended design, but it is not the requested diagnosis of Increase completion length to improve medical authority.
B:This limitation concerns Disable all source links and let the model reason unaudited; it does not diagnose the proposed Increase completion length to improve medical authority approach.
C:This limitation concerns Treat any confident generated dosage as approved clinical guidance; it does not diagnose the proposed Increase completion length to improve medical authority approach.
