AWS Certified AI Practitioner AIF-C01 Practice Test 37
Question 360
An image model should avoid including brand logos in generated artwork. 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: Set the output token count to one as a logo filter
- A system inventory lists this competing approach as active: Mention the prohibited logo as a desired featured element
- Human review confirms prohibited logos are absent from sample images
- A reviewed design record approves this competing approach: Assume only a model's training history prevents every unwanted brand
Explanation
Correct Answer
C:Human review confirms prohibited logos are absent from sample images. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Length settings are not semantic content filters A component invocation does not validate the necessary end-to-end outcome.
B:Positive instructions can encourage the excluded object An active setting cannot independently prove the requested business outcome.
D:Training provenance alone does not enforce prompt-specific exclusions Design approval is not evidence that the alternative meets the actual requirement.
Question 361
A firm must disclose that customers are interacting with AI and provide escalation. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- Users can identify the AI and reach a human when the result is disputed
- A reviewed design record approves this competing approach: Require users to accept all outputs without correction
- A system inventory lists this competing approach as active: Hide the model identity to make replies seem human
- A component test invokes this alternate approach successfully: Use only automated satisfaction guesses without user recourse
Explanation
Correct Answer
A:Users can identify the AI and reach a human when the result is disputed. This checks the business result rather than just a configured feature.
Incorrect Answers
B:No contest or escalation path weakens trust Design approval is not evidence that the alternative meets the actual requirement.
C:Deception removes user choice and feedback An active setting cannot independently prove the requested business outcome.
D:Predicted sentiment does not let users contest outcomes A component invocation does not validate the necessary end-to-end outcome.
Question 362
A regulated firm must document a model's intended use, limitations and risk rating. Which option BEST meets the business requirement?
- Use billing invoices as the complete model-risk record
- Maintain an Amazon SageMaker Model Card for that model
- Assume a saved model weight file documents model lineage and risk
- Treat a dataset location as the entire model documentation
Explanation
Correct Answer
B:Model Cards capture model metadata, intended uses and risk considerations The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Cost reports omit intended uses and technical limitations This prevents the approach from meeting the stated requirement.
C:Weights alone do not describe governance decisions This prevents the approach from meeting the stated requirement.
D:A storage path cannot capture evaluation and intended use This prevents the approach from meeting the stated requirement.
Question 363
An insurer already using SageMaker Clarify needs feature-level explanations of an existing tabular risk model. Before approving a rollout, an architect suggests Use S3 object version IDs as causal model explanations. Which mismatch should be documented for this precise design?
- Encryption manages key protection rather than attribution
- Activity logs do not calculate feature attribution
- Clarify supports attribution and bias analysis for supported models
- Object versions do not explain predicted scores
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Use S3 object version IDs as causal model explanations. Object versions do not explain predicted scores
Incorrect Answers
A:This limitation concerns Use KMS encryption keys as explanations; it does not diagnose the proposed Use S3 object version IDs as causal model explanations approach.
B:This limitation concerns Use CloudTrail API logs as model-feature explanations; it does not diagnose the proposed Use S3 object version IDs as causal model explanations approach.
C:Clarify supports attribution and bias analysis for supported models This explains the recommended design, but it is not the requested diagnosis of Use S3 object version IDs as causal model explanations.
Question 364
A bank asks how to communicate performance limitations for a fraud model. A pilot uses this approach instead: Publish only a marketing statement about AI accuracy. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Availability does not describe failure modes or bias
- Marketing copy omits measured limitations and assumptions
- Transparent documentation lets users understand where output is reliable
- Weights cannot serve as a plain-language limitation record
Explanation
Correct Answer
B:This is the actual failure mode of the proposed approach: Publish only a marketing statement about AI accuracy. Marketing copy omits measured limitations and assumptions
Incorrect Answers
A:This limitation concerns Provide only a service uptime chart as model quality disclosure; it does not diagnose the proposed Publish only a marketing statement about AI accuracy approach.
C:Transparent documentation lets users understand where output is reliable This explains the recommended design, but it is not the requested diagnosis of Publish only a marketing statement about AI accuracy.
D:This limitation concerns Publish only the model's hidden weights; it does not diagnose the proposed Publish only a marketing statement about AI accuracy approach.
Question 365
An external chat assistant sometimes returns insulting or hateful language. 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 only a vector database index as the content safety filter
- Test harmful prompts are blocked or handled according to policy
- A reviewed design record approves this competing approach: Rely only on a polite system persona without safety testing
- A component test invokes this alternate approach successfully: Change IAM billing permissions to remove toxic language
Explanation
Correct Answer
B:Test harmful prompts are blocked or handled according to policy. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Retrieval search is not an output toxicity control An active setting cannot independently prove the requested business outcome.
C:Prompt wording alone is not a complete moderation control Design approval is not evidence that the alternative meets the actual requirement.
D:Billing authorization does not inspect generated text A component invocation does not validate the necessary end-to-end outcome.
Question 366
A high-impact customer eligibility model has borderline cases needing a human decision. A pilot uses this approach instead: Automatically approve every high-confidence output without defined controls. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Toxicity filtering does not replace human eligibility judgment
- Reviewers can inspect uncertain outputs before critical action
- Faster inference is not evidence of acceptable decision risk
- Confidence alone may not meet policy for sensitive outcomes
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Automatically approve every high-confidence output without defined controls. Confidence alone may not meet policy for sensitive outcomes
Incorrect Answers
A:This limitation concerns Treat a guardrail toxicity filter as the final eligibility reviewer; it does not diagnose the proposed Automatically approve every high-confidence output without defined controls approach.
B:Reviewers can inspect uncertain outputs before critical action This explains the recommended design, but it is not the requested diagnosis of Automatically approve every high-confidence output without defined controls.
C:This limitation concerns Use lower latency as proof that human review is unnecessary; it does not diagnose the proposed Automatically approve every high-confidence output without defined controls approach.
Question 367
A deployed predictor starts receiving different input distributions from the data used for validation. Which option BEST meets the business requirement?
- Monitor data drift and model quality using supported monitoring capabilities
- Retrain continuously without tracking data or performance changes
- Measure only S3 storage growth as model drift
- Assume a successful deployment permanently guarantees accuracy
Explanation
Correct Answer
A:Production input changes can degrade predictive behavior The specified scenario makes this the most appropriate approach.
Incorrect Answers
B:Uncontrolled retraining can introduce new risk This prevents the approach from meeting the stated requirement.
C:Storage bytes do not diagnose feature distribution change This prevents the approach from meeting the stated requirement.
D:Deployment state says nothing about future distribution shift This prevents the approach from meeting the stated requirement.
Question 368
An AI assistant's refusals frustrate legitimate users while missing some unsafe requests. A pilot uses this approach instead: Set all filters maximally restrictive without usability testing. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Raw block count does not reveal missed harms or unjustified refusals
- No filtering exposes users to prohibited content
- Safety requires balancing blocked harmful output with useful allowed output
- Overblocking can prevent legitimate business activity
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Set all filters maximally restrictive without usability testing. Overblocking can prevent legitimate business activity
Incorrect Answers
A:This limitation concerns Judge only the total number of blocked prompts; it does not diagnose the proposed Set all filters maximally restrictive without usability testing approach.
B:This limitation concerns Turn all safety filters off to eliminate refusals; it does not diagnose the proposed Set all filters maximally restrictive without usability testing approach.
C:Safety requires balancing blocked harmful output with useful allowed output This explains the recommended design, but it is not the requested diagnosis of Set all filters maximally restrictive without usability testing.
Question 369
A bank asks how to communicate performance limitations for a fraud model. During a design review, someone proposes Publish only the model's hidden weights. Which technical objection is specifically relevant to this proposal?
- Weights cannot serve as a plain-language limitation record
- Marketing copy omits measured limitations and assumptions
- Availability does not describe failure modes or bias
- Transparent documentation lets users understand where output is reliable
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Publish only the model's hidden weights. Weights cannot serve as a plain-language limitation record
Incorrect Answers
B:This limitation concerns Publish only a marketing statement about AI accuracy; it does not diagnose the proposed Publish only the model's hidden weights approach.
C:This limitation concerns Provide only a service uptime chart as model quality disclosure; it does not diagnose the proposed Publish only the model's hidden weights approach.
D:Transparent documentation lets users understand where output is reliable This explains the recommended design, but it is not the requested diagnosis of Publish only the model's hidden weights.
Question 370
A model offers high accuracy but opaque reasons for high-stakes decisions. A pilot uses this approach instead: Choose highest accuracy without assessing auditability. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Interpretability alone does not 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
D:This is the actual failure mode of the proposed approach: Choose highest accuracy without assessing auditability. Pure accuracy can fail explanation obligations
Incorrect Answers
A:This limitation concerns Choose the simplest model regardless of accuracy or safety; it does not diagnose the proposed Choose highest accuracy without assessing auditability 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 Choose highest accuracy without assessing auditability.
C:This limitation concerns Use model popularity as the risk-adjusted decision; it does not diagnose the proposed Choose highest accuracy without assessing auditability approach.
