Amazon AWS Certified AI Practitioner AIF-C01 Practice Test 39
Question 380
The initial scenario was: A chatbot occasionally responds to hostile prompts with unsafe instructions despite a written 'be safe' guideline. The original control is already in place. Now the app integrates an external action tool. Which ADDITIONAL action BEST addresses the new condition?
- Add tool-level authorization and action safety checks
- A red-team suite exercises policy-violating requests and records failures
- Test adversarial prompts and add layered content-safety controls
- Conclude a single benign prompt proves resistance to jailbreaks
Explanation
Correct Answer
A:The app integrates an external action tool. The appropriate added action is: Add tool-level authorization and action safety checks.
Incorrect Answers
B:This describes evidence for the original approach, not a new action: A red-team suite exercises policy-violating requests and records failures.
C:This was the original control and is already implemented. Safety depends on actual enforcement and monitoring, not a slogan.
D:This original shortcut remains unsuitable: One successful test does not establish robustness to adversarial variants.
Question 381
A lender improves average prediction accuracy but significantly worsens false-negative rates for a minority customer group. Which TWO statements jointly identify the best initial approach and an observable verification result? (Select TWO.)
- Compare subgroup metrics and fairness impacts before deployment
- Require citation verification and checks against trusted sources
- Ignore subgroup deterioration because the global average rose
- A release review includes per-group error rates and corrective actions
- Add appropriate explanation methods and documented human recourse
Explanation
Correct Answers
A:Correct approach. Accuracy improvements can conflict with equitable treatment across groups. This directly serves the stated situation: A lender improves average prediction accuracy but significantly worsens false-negative rates for a minority customer group.
D:Correct operating evidence. Observing a release review includes per-group error rates and corrective actions verifies this approach instead of relying on unperformed plans.
Incorrect Answers
B:This addresses a different scenario: An assistant generates plausible references to scientific studies that do not exist. It does not constitute the requested approach or operating result for the current case.
C:A favorable global score can mask unequal error rates. The suggested shortcut would not meet the present requirement.
E:This would be appropriate if the separate requirement arose: The team later wants interpretable reasons for adverse decisions. That is not the present condition.
Question 382
A lender improves average prediction accuracy but significantly worsens false-negative rates for a minority customer group. A colleague suggests: Ignore subgroup deterioration because the global average rose. Which finding BEST explains why the suggestion should be rejected?
- Model scale can increase compute demand without proportional task benefit
- A favorable global score can mask unequal error rates
- The shortcut is safe because its terminology is familiar to the team
- Fluent source-looking text does not make a citation real
Explanation
Correct Answer
B:A favorable global score can mask unequal error rates. This is the direct defect in the proposed action: Ignore subgroup deterioration because the global average rose.
Incorrect Answers
A:This concern belongs to another situation: A team wants to compare how much energy and infrastructure a larger foundation model consumes relative to a smaller suitable model.
C:Familiar wording is not proof that an action is appropriate. The actual defect is: A favorable global score can mask unequal error rates.
D:This finding is associated with a different problem: An assistant generates plausible references to scientific studies that do not exist.
Question 383
A lender improves average prediction accuracy but significantly worsens false-negative rates for a minority customer group. After implementing Compare subgroup metrics and fairness impacts before deployment, which observation BEST verifies the intended result?
- The team schedules a future test of whether a release review includes per-group error rates and corrective actions, but has no result
- Test answers include only validated source references or uncertainty statements
- A release review includes per-group error rates and corrective actions
- The team approves Compare subgroup metrics and fairness impacts before deployment but does not run an operational test
Explanation
Correct Answer
C:The observed result, A release review includes per-group error rates and corrective actions, provides direct evidence. Accuracy improvements can conflict with equitable treatment across groups.
Incorrect Answers
A:An unperformed future check is not evidence. Verify the actual outcome: A release review includes per-group error rates and corrective actions.
B:This observation would be evidence for another case: An assistant generates plausible references to scientific studies that do not exist, not the current scenario.
D:The required observable condition is: A release review includes per-group error rates and corrective actions.
Question 384
The initial scenario was: A lender improves average prediction accuracy but significantly worsens false-negative rates for a minority customer group. The original control is already in place. Now the team later wants interpretable reasons for adverse decisions. Which ADDITIONAL action BEST addresses the new condition?
- Compare subgroup metrics and fairness impacts before deployment
- A release review includes per-group error rates and corrective actions
- Ignore subgroup deterioration because the global average rose
- Add appropriate explanation methods and documented human recourse
Explanation
Correct Answer
D:The team later wants interpretable reasons for adverse decisions. The appropriate added action is: Add appropriate explanation methods and documented human recourse.
Incorrect Answers
A:This was the original control and is already implemented. Accuracy improvements can conflict with equitable treatment across groups.
B:This describes evidence for the original approach, not a new action: A release review includes per-group error rates and corrective actions.
C:This original shortcut remains unsuitable: A favorable global score can mask unequal error rates. It is not a valid solution to the changed requirement.
Question 385
An assistant generates plausible references to scientific studies that do not exist. Which option BEST meets the stated AI/ML or AWS requirement?
- Use grounded retrieval with source-linked passages
- Present invented bibliography entries as confirmed evidence
- Require citation verification and checks against trusted sources
- Include sustainability and resource use in model selection
Explanation
Correct Answer
C:Hallucinated attribution undermines truthfulness and user trust. In this situation, choose: Require citation verification and checks against trusted sources.
Incorrect Answers
A:This is appropriate only after a distinct change: The organization builds a curated research library. It is not the best initial response to an assistant generates plausible references to scientific studies that do not exist.
B:Fluent source-looking text does not make a citation real. The proposal, Present invented bibliography entries as confirmed evidence, does not satisfy this need.
D:This can be valid for a different task: A team wants to compare how much energy and infrastructure a larger foundation model consumes relative to a smaller suitable model. The current request is different and needs the selected approach.
Question 386
An assistant generates plausible references to scientific studies that do not exist. A colleague suggests: Present invented bibliography entries as confirmed evidence. Which finding BEST explains why the suggestion should be rejected?
- Model scale can increase compute demand without proportional task benefit
- The shortcut is safe because its terminology is familiar to the team
- Fluent source-looking text does not make a citation real
- Encryption protects data and does not explain a model's decisions
Explanation
Correct Answer
C:Fluent source-looking text does not make a citation real. This is the direct defect in the proposed action: Present invented bibliography entries as confirmed evidence.
Incorrect Answers
A:This finding is associated with a different problem: A team wants to compare how much energy and infrastructure a larger foundation model consumes relative to a smaller suitable model.
B:Familiar wording is not proof that an action is appropriate. The actual defect is: Fluent source-looking text does not make a citation real.
D:This concern belongs to another situation: A credit-risk model's feature contributions are opaque and reviewers want to examine potentially discriminatory input influence.
Question 387
An assistant generates plausible references to scientific studies that do not exist. After implementing Require citation verification and checks against trusted sources, which observation BEST verifies the intended result?
- Model selection documents quality and compute tradeoffs for tested workloads
- The team schedules a future test of whether test answers include only validated source references or uncertainty statements, but has no result
- The team approves Require citation verification and checks against trusted sources but does not run an operational test
- Test answers include only validated source references or uncertainty statements
Explanation
Correct Answer
D:The observed result, Test answers include only validated source references or uncertainty statements, provides direct evidence. Hallucinated attribution undermines truthfulness and user trust.
Incorrect Answers
A:This observation would be evidence for another case: A team wants to compare how much energy and infrastructure a larger foundation model consumes relative to a smaller suitable model, not the current scenario.
B:Verify the actual outcome: Test answers include only validated source references or uncertainty statements.
C:The required observable condition is: Test answers include only validated source references or uncertainty statements.
Question 388
The initial scenario was: An assistant generates plausible references to scientific studies that do not exist. The original control is already in place. Now the organization builds a curated research library. Which ADDITIONAL action BEST addresses the new condition?
- Present invented bibliography entries as confirmed evidence
- Require citation verification and checks against trusted sources
- Use grounded retrieval with source-linked passages
- Test answers include only validated source references or uncertainty statements
Explanation
Correct Answer
C:The organization builds a curated research library. The appropriate added action is: Use grounded retrieval with source-linked passages.
Incorrect Answers
A:This original shortcut remains unsuitable: Fluent source-looking text does not make a citation real.
B:This was the original control and is already implemented. Hallucinated attribution undermines truthfulness and user trust.
D:This describes evidence for the original approach, not a new action: Test answers include only validated source references or uncertainty statements.
Question 389
A team wants to compare how much energy and infrastructure a larger foundation model consumes relative to a smaller suitable model. Which option BEST meets the stated AI/ML or AWS requirement?
- Assume the largest model always has the best environmental outcome
- Include sustainability and resource use in model selection
- Use Amazon SageMaker Clarify for applicable bias and explainability analysis
- Rebalance sustainability against demonstrated quality and safety requirements
Explanation
Correct Answer
B:Responsible AI considers resource efficiency alongside capability and cost. In this situation, choose: Include sustainability and resource use in model selection.
Incorrect Answers
A:Model scale can increase compute demand without proportional task benefit. The proposal, Assume the largest model always has the best environmental outcome, does not satisfy this need.
C:This can be valid for a different task: A credit-risk model's feature contributions are opaque and reviewers want to examine potentially discriminatory input influence. The current request is different and needs the selected approach.
D:This is appropriate only after a distinct change: A small model performs poorly on the most important tasks. It is not the best initial response to a team wants to compare how much energy and infrastructure a larger foundation model consumes relative to a smaller suitable model.
Question 390
A team wants to compare how much energy and infrastructure a larger foundation model consumes relative to a smaller suitable model. A colleague suggests: Assume the largest model always has the best environmental outcome. Which finding BEST explains why the suggestion should be rejected?
- Shifting populations can invalidate historical quality conclusions
- Model scale can increase compute demand without proportional task benefit
- The shortcut is safe because its terminology is familiar to the team
- Encryption protects data and does not explain a model's decisions
Explanation
Correct Answer
B:Model scale can increase compute demand without proportional task benefit. This is the direct defect in the proposed action: Assume the largest model always has the best environmental outcome.
Incorrect Answers
A:This concern belongs to another situation: A classification system performs well in testing but becomes less accurate after production data patterns shift.
C:Familiar wording is not proof that an action is appropriate. The actual defect is: Model scale can increase compute demand without proportional task benefit.
D:This finding is associated with a different problem: A credit-risk model's feature contributions are opaque and reviewers want to examine potentially discriminatory input influence.