AWS Certified AI Practitioner AIF-C01 Practice Test 21
Question 200
A product owner is choosing between text-only and image-capable models for photo-assisted replies. Which option BEST meets the business requirement?
- Use text-only just because it is a foundation model
- Select a model with the required image input modality
- Assume language support automatically includes photos
- Select by model popularity rather than interface capabilities
Explanation
Correct Answer
B:Models differ in supported input/output types The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:FM status does not imply image input This prevents the approach from meeting the stated requirement.
C:Multilingual and multimodal support are independent This prevents the approach from meeting the stated requirement.
D:Popularity cannot guarantee image understanding This prevents the approach from meeting the stated requirement.
Question 201
A fixed-label fraud predictor is being replaced by a much larger text-generating FM without justification. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- A reviewed design record approves this competing approach: Generate synthetic fraud illustrations as the decision itself
- Compare fraud precision, recall, latency and cost for appropriate models
- A component test invokes this alternate approach successfully: Choose the largest FM without comparative tests
- A system inventory lists this competing approach as active: Require all AI workloads to generate free text
Explanation
Correct Answer
B:Compare fraud precision, recall, latency and cost for appropriate models. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Image generation does not provide a reliable binary score Design approval is not evidence that the alternative meets the actual requirement.
C:Size does not guarantee a fit for tabular fraud decisions A component invocation does not validate the necessary end-to-end outcome.
D:Prediction is valid AI without content generation An active setting cannot independently prove the requested business outcome.
Question 202
A retailer wants to show whether an assistant increases useful customer outcomes. Which option BEST meets the business requirement?
- Measure resolution, satisfaction and cost per successful interaction
- Treat parameter count as net revenue
- Use model training accuracy as the sole business metric
- Use tokens generated per second as the ROI metric
Explanation
Correct Answer
A:Business value requires outcome measures not only token throughput The specified scenario makes this the most appropriate approach.
Incorrect Answers
B:Model size is not a financial outcome This prevents the approach from meeting the stated requirement.
C:Training accuracy does not capture customer conversations This prevents the approach from meeting the stated requirement.
D:Speed alone does not measure customer resolution This prevents the approach from meeting the stated requirement.
Question 203
Several specialized agents must research claims, calculate a payment and request final human approval. A pilot uses this approach instead: Grant every agent unrestricted payout permissions. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Retrieval alone does not authorize or execute payment
- Coordinated agents can combine outputs while preserving authority checks
- Unrestricted write access bypasses human approval
- Disconnected assistants cannot reliably compose the process
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Grant every agent unrestricted payout permissions. Unrestricted write access bypasses human approval
Incorrect Answers
A:This limitation concerns Use embeddings search as a final claim settlement; it does not diagnose the proposed Grant every agent unrestricted payout permissions approach.
B:Coordinated agents can combine outputs while preserving authority checks This explains the recommended design, but it is not the requested diagnosis of Grant every agent unrestricted payout permissions.
D:This limitation concerns Use isolated chatbots with no handoff state; it does not diagnose the proposed Grant every agent unrestricted payout permissions approach.
Question 204
A help center must choose an FM for short multilingual support replies. A pilot uses this approach instead: Always choose the largest parameter count. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Size alone does not prove practical fit
- Price is unable to compensate for missing languages
- Real workload requirements determine model fit
- Long output capability does not establish short multilingual accuracy
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Always choose the largest parameter count. Size alone does not prove practical fit
Incorrect Answers
B:This limitation concerns Choose cheapest regardless of language support; it does not diagnose the proposed Always choose the largest parameter count approach.
C:Real workload requirements determine model fit This explains the recommended design, but it is not the requested diagnosis of Always choose the largest parameter count.
D:This limitation concerns Select by maximum output tokens alone; it does not diagnose the proposed Always choose the largest parameter count approach.
Question 205
An insurance assistant must interpret an uploaded photo together with a text instruction. Which option BEST meets the business requirement?
- Use audio transcription as image understanding
- Select a multimodal model accepting image and text
- Use a filename-based regression score as the visual interpretation
- Use text-only inference without a visual description
Explanation
Correct Answer
B:Multimodal input processing covers both evidence types The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Speech recognition does not process picture content This prevents the approach from meeting the stated requirement.
C:File names are not evidence of damage in pixels This prevents the approach from meeting the stated requirement.
D:Text-only input cannot directly inspect the photo This prevents the approach from meeting the stated requirement.
Question 206
An assistant gives convincing but invented policy clauses. Which option BEST meets the business requirement?
- Increase sampling randomness to ensure correct facts
- Treat unsupported claims as hallucination risk and verify against source documents
- Treat output length as a correctness certificate
- Assume fluent wording guarantees factual accuracy
Explanation
Correct Answer
B:Fluent LLM output can still be false The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Higher randomness cannot create missing evidence This prevents the approach from meeting the stated requirement.
C:Long responses may contain unsupported content This prevents the approach from meeting the stated requirement.
D:Style is not factual evidence This prevents the approach from meeting the stated requirement.
Question 207
A data scientist wants to experiment, train and deploy a custom time-series predictor. Before approving a rollout, an architect suggests Use AWS Config as feature engineering and model hosting. Which mismatch should be documented for this precise design?
- Guardrails filter inputs and outputs rather than train models
- SageMaker AI supports the broader ML lifecycle
- Config audits resource settings, not ML training
- Polly synthesizes speech rather than training arbitrary predictors
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Use AWS Config as feature engineering and model hosting. Config audits resource settings, not ML training
Incorrect Answers
A:This limitation concerns Use Bedrock Guardrails to perform custom training; it does not diagnose the proposed Use AWS Config as feature engineering and model hosting approach.
B:SageMaker AI supports the broader ML lifecycle This explains the recommended design, but it is not the requested diagnosis of Use AWS Config as feature engineering and model hosting.
D:This limitation concerns Use Polly as the general model-training platform; it does not diagnose the proposed Use AWS Config as feature engineering and model hosting approach.
Question 208
An assistant gives convincing but invented policy clauses. During a design review, someone proposes Increase sampling randomness to ensure correct facts. Which technical objection is specifically relevant to this proposal?
- Fluent LLM output can still be false
- Style is not factual evidence
- Higher randomness cannot create missing evidence
- Long responses may contain unsupported content
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Increase sampling randomness to ensure correct facts. Higher randomness cannot create missing evidence
Incorrect Answers
A:Fluent LLM output can still be false This explains the recommended design, but it is not the requested diagnosis of Increase sampling randomness to ensure correct facts.
B:This limitation concerns Assume fluent wording guarantees factual accuracy; it does not diagnose the proposed Increase sampling randomness to ensure correct facts approach.
D:This limitation concerns Treat output length as a correctness certificate; it does not diagnose the proposed Increase sampling randomness to ensure correct facts approach.
Question 209
An agent must access approved external APIs through a common tool interface. During a design review, someone proposes Use vector relevance as permission to transfer money. Which technical objection is specifically relevant to this proposal?
- Similarity is not authorization
- Breaking text into tokens does not authenticate API calls
- Prompt text cannot establish secure tool registration
- MCP standardizes tool connectivity but does not grant permission automatically
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Use vector relevance as permission to transfer money. Similarity is not authorization
Incorrect Answers
B:This limitation concerns Use tokenization as the API authorization mechanism; it does not diagnose the proposed Use vector relevance as permission to transfer money approach.
C:This limitation concerns Assume mentioning a tool in a prompt makes it callable; it does not diagnose the proposed Use vector relevance as permission to transfer money approach.
D:MCP standardizes tool connectivity but does not grant permission automatically This explains the recommended design, but it is not the requested diagnosis of Use vector relevance as permission to transfer money.
Question 210
Several specialized agents must research claims, calculate a payment and request final human approval. 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: Grant every agent unrestricted payout permissions
- A component test invokes this alternate approach successfully: Use isolated chatbots with no handoff state
- A reviewed design record approves this competing approach: Use embeddings search as a final claim settlement
- Audit traces show subtask results and an approval gate before payout
Explanation
Correct Answer
D:Audit traces show subtask results and an approval gate before payout. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Unrestricted write access bypasses human approval An active setting cannot independently prove the requested business outcome.
B:Disconnected assistants cannot reliably compose the process A component invocation does not validate the necessary end-to-end outcome.
C:Retrieval alone does not authorize or execute payment Design approval is not evidence that the alternative meets the actual requirement.
