AWS Certified AI Practitioner AIF-C01 Practice Test 41
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.
Question 401
A credit-screening model performs well overall but misses qualified applicants in one protected group. During a design review, someone proposes Judge fairness solely by model parameter count. Which technical objection is specifically relevant to this proposal?
- Size does not measure differential impact
- A high average score can coexist with harmful group-level errors
- Aggregate metrics can hide disparities across populations
- Without appropriate controlled subgroup information the disparity may remain hidden
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Judge fairness solely by model parameter count. Size does not measure differential impact
Incorrect Answers
B:This limitation concerns Use overall accuracy as the only fairness measure; it does not diagnose the proposed Judge fairness solely by model parameter count approach.
C:Aggregate metrics can hide disparities across populations This explains the recommended design, but it is not the requested diagnosis of Judge fairness solely by model parameter count.
D:This limitation concerns Remove demographic evaluation fields before measuring disparities; it does not diagnose the proposed Judge fairness solely by model parameter count approach.
Question 402
A customer-facing medical information assistant may invent unsafe dosage recommendations. A pilot uses this approach instead: Treat any confident generated dosage as approved clinical guidance. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Longer advice is not clinical validation
- High-stakes unsupported medical claims need source checking and human control
- Model confidence cannot establish medical safety
- Removing provenance increases the chance of undetected errors
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Treat any confident generated dosage as approved clinical guidance. Model confidence cannot establish medical safety
Incorrect Answers
A:This limitation concerns Increase completion length to improve medical authority; it does not diagnose the proposed Treat any confident generated dosage as approved clinical guidance approach.
B:High-stakes unsupported medical claims need source checking and human control This explains the recommended design, but it is not the requested diagnosis of Treat any confident generated dosage as approved clinical guidance.
D:This limitation concerns Disable all source links and let the model reason unaudited; it does not diagnose the proposed Treat any confident generated dosage as approved clinical guidance approach.
Question 403
A language model is being adapted for a multilingual community with underrepresented dialects. Which option BEST meets the business requirement?
- Train only on dominant dialect samples and assume transfer is perfect
- Use only one high-performing standard-language benchmark
- Include diverse representative data and evaluate performance by dialect
- Increase output tokens as the entire inclusion strategy
Explanation
Correct Answer
C:Inclusive datasets and subgroup testing help identify representation gaps The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Missing dialect examples can harm the omitted users This prevents the approach from meeting the stated requirement.
B:A single dialect benchmark cannot prove broader coverage This prevents the approach from meeting the stated requirement.
D:More generated words cannot add missing training representation This prevents the approach from meeting the stated requirement.
Question 404
A language model is being adapted for a multilingual community with underrepresented dialects. During a design review, someone proposes Use only one high-performing standard-language benchmark. Which technical objection is specifically relevant to this proposal?
- Missing dialect examples can harm the omitted users
- More generated words cannot add missing training representation
- Inclusive datasets and subgroup testing help identify representation gaps
- A single dialect benchmark cannot prove broader coverage
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Use only one high-performing standard-language benchmark. A single dialect benchmark cannot prove broader coverage
Incorrect Answers
A:This limitation concerns Train only on dominant dialect samples and assume transfer is perfect; it does not diagnose the proposed Use only one high-performing standard-language benchmark approach.
B:This limitation concerns Increase output tokens as the entire inclusion strategy; it does not diagnose the proposed Use only one high-performing standard-language benchmark approach.
C:Inclusive datasets and subgroup testing help identify representation gaps This explains the recommended design, but it is not the requested diagnosis of Use only one high-performing standard-language benchmark.
Question 405
A firm must disclose that customers are interacting with AI and provide escalation. A pilot uses this approach instead: Hide the model identity to make replies seem human. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Predicted sentiment does not let users contest outcomes
- Deception removes user choice and feedback
- Human-centered design supports informed use and correction of mistakes
- No contest or escalation path weakens trust
Explanation
Correct Answer
B:This is the actual failure mode of the proposed approach: Hide the model identity to make replies seem human. Deception removes user choice and feedback
Incorrect Answers
A:This limitation concerns Use only automated satisfaction guesses without user recourse; it does not diagnose the proposed Hide the model identity to make replies seem human approach.
C:Human-centered design supports informed use and correction of mistakes This explains the recommended design, but it is not the requested diagnosis of Hide the model identity to make replies seem human.
D:This limitation concerns Require users to accept all outputs without correction; it does not diagnose the proposed Hide the model identity to make replies seem human approach.
Question 406
An external chat assistant sometimes returns insulting or hateful language. Before approving a rollout, an architect suggests Rely only on a polite system persona without safety testing. Which mismatch should be documented for this precise design?
- Retrieval search is not an output toxicity control
- Billing authorization does not inspect generated text
- Prompt wording alone is not a complete moderation control
- Guardrails can help filter unwanted model inputs and outputs
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Rely only on a polite system persona without safety testing. Prompt wording alone is not a complete moderation control
Incorrect Answers
A:This limitation concerns Use only a vector database index as the content safety filter; it does not diagnose the proposed Rely only on a polite system persona without safety testing approach.
B:This limitation concerns Change IAM billing permissions to remove toxic language; it does not diagnose the proposed Rely only on a polite system persona without safety testing approach.
D:Guardrails can help filter unwanted model inputs and outputs This explains the recommended design, but it is not the requested diagnosis of Rely only on a polite system persona without safety testing.
Question 407
A model offers high accuracy but opaque reasons for high-stakes decisions. 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: Choose the simplest model regardless of accuracy or safety
- A system inventory lists this competing approach as active: Choose highest accuracy without assessing auditability
- A reviewed design record approves this competing approach: Use model popularity as the risk-adjusted decision
- A governance review documents the selected performance-explainability tradeoff
Explanation
Correct Answer
D:A governance review documents the selected performance-explainability tradeoff. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Interpretability alone does not guarantee acceptable outcomes A component invocation does not validate the necessary end-to-end outcome.
B:Pure accuracy can fail explanation obligations An active setting cannot independently prove the requested business outcome.
C:Adoption counts do not measure explainability or model reliability Design approval is not evidence that the alternative meets the actual requirement.
Question 408
An analyst needs a model explanation that business users can understand without revealing training secrets. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- User testing confirms the explanation is understandable and privacy-safe
- A system inventory lists this competing approach as active: Reveal private training rows as the explanation
- A reviewed design record approves this competing approach: Provide a billing receipt instead of decision rationale
- A component test invokes this alternate approach successfully: Use only neural network layer counts for every decision
Explanation
Correct Answer
A:User testing confirms the explanation is understandable and privacy-safe. This checks the business result rather than just a configured feature.
Incorrect Answers
B:Exposing source records can violate privacy and still be confusing An active setting cannot independently prove the requested business outcome.
C:Charges do not establish why the model responded Design approval is not evidence that the alternative meets the actual requirement.
D:Architecture size does not explain a particular prediction A component invocation does not validate the necessary end-to-end outcome.
Question 409
A deployed predictor starts receiving different input distributions from the data used for validation. 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: Assume a successful deployment permanently guarantees accuracy
- A reviewed design record approves this competing approach: Measure only S3 storage growth as model drift
- A monitored baseline detects distribution shifts and triggers a review
- A component test invokes this alternate approach successfully: Retrain continuously without tracking data or performance changes
Explanation
Correct Answer
C:A monitored baseline detects distribution shifts and triggers a review. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Deployment state says nothing about future distribution shift An active setting cannot independently prove the requested business outcome.
B:Storage bytes do not diagnose feature distribution change Design approval is not evidence that the alternative meets the actual requirement.
D:Uncontrolled retraining can introduce new risk A component invocation does not validate the necessary end-to-end outcome.
Question 410
An external chat assistant sometimes returns insulting or hateful language. A pilot uses this approach instead: Use only a vector database index as the content safety filter. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Guardrails can help filter unwanted model inputs and outputs
- Prompt wording alone is not a complete moderation control
- Billing authorization does not inspect generated text
- Retrieval search is not an output toxicity control
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Use only a vector database index as the content safety filter. Retrieval search is not an output toxicity control
Incorrect Answers
A:Guardrails can help filter unwanted model inputs and outputs This explains the recommended design, but it is not the requested diagnosis of Use only a vector database index as the content safety filter.
B:This limitation concerns Rely only on a polite system persona without safety testing; it does not diagnose the proposed Use only a vector database index as the content safety filter approach.
C:This limitation concerns Change IAM billing permissions to remove toxic language; it does not diagnose the proposed Use only a vector database index as the content safety filter approach.
