AWS Certified AI Practitioner AIF-C01 Practice Test 11
Question 100
A product manager confuses generative AI with ordinary binary prediction of account fraud. 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: Describe deep learning as unrelated to machine learning
- A component test invokes this alternate approach successfully: Assume a language model is required for every tabular classification
- Review the business output requirement and verify whether it is a label or newly generated content
- A system inventory lists this competing approach as active: Treat every neural network's output as necessarily a generated paragraph
Explanation
Correct Answer
C:Review the business output requirement and verify whether it is a label or newly generated content. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Deep learning is a family of machine learning methods Design approval is not evidence that the alternative meets the actual requirement.
B:A tabular classifier may be sufficient and cheaper for labeled fixed classes A component invocation does not validate the necessary end-to-end outcome.
D:Neural networks can be used for classification and regression as well as generation An active setting cannot independently prove the requested business outcome.
Question 101
A policy summarizer generates creative off-topic responses. During a design review, someone proposes Remove all source context. Which technical objection is specifically relevant to this proposal?
- More generated tokens do not ensure relevance
- High randomness increases diversity rather than factual assurance
- Lower sampling randomness can reduce variation but does not replace grounding
- Lack of evidence increases unsupported-answer risk
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Remove all source context. Lack of evidence increases unsupported-answer risk
Incorrect Answers
A:This limitation concerns Increase output length without evaluating quality; it does not diagnose the proposed Remove all source context approach.
B:This limitation concerns Maximize temperature to guarantee faithfulness; it does not diagnose the proposed Remove all source context approach.
C:Lower sampling randomness can reduce variation but does not replace grounding This explains the recommended design, but it is not the requested diagnosis of Remove all source context.
Question 102
A team is selecting how to serve a production FM under predictable sustained load. Which option BEST meets the business requirement?
- Always buy dedicated throughput for a one-off experiment
- Assume on-demand never has rate or quota limits
- Compare on-demand inference with available provisioned-throughput options
- Train a new model on every request to guarantee throughput
Explanation
Correct Answer
C:Capacity choice should reflect sustained traffic, latency and pricing The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Reserved capacity can waste money under sporadic load This prevents the approach from meeting the stated requirement.
B:On-demand services can still have limits and performance constraints This prevents the approach from meeting the stated requirement.
D:Training per inference is not a viable serving strategy This prevents the approach from meeting the stated requirement.
Question 103
A policy summarizer generates creative off-topic responses. 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: Increase output length without evaluating quality
- A system inventory lists this competing approach as active: Maximize temperature to guarantee faithfulness
- Repeated cases show fewer off-topic responses without lost coverage
- A component test invokes this alternate approach successfully: Remove all source context
Explanation
Correct Answer
C:Repeated cases show fewer off-topic responses without lost coverage. This checks the business result rather than just a configured feature.
Incorrect Answers
A:More generated tokens do not ensure relevance Design approval is not evidence that the alternative meets the actual requirement.
B:High randomness increases diversity rather than factual assurance An active setting cannot independently prove the requested business outcome.
D:Lack of evidence increases unsupported-answer risk A component invocation does not validate the necessary end-to-end outcome.
Question 104
A customer expects identical exact wording for every creative-answer request. Which option BEST meets the business requirement?
- Assume longer prompts alone guarantee equality
- Add an embedding database to force identical prose
- Explain generative nondeterminism and check required output constraints
- Promise byte-identical text from every call
Explanation
Correct Answer
C:Sampling and model conditions can vary completions The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Length is not a reproducibility guarantee This prevents the approach from meeting the stated requirement.
B:Retrieval does not force identical generation This prevents the approach from meeting the stated requirement.
D:Generative outputs are not universally deterministic This prevents the approach from meeting the stated requirement.
Question 105
A support chatbot drops critical policy sections when every long document is pasted into its prompt. During a design review, someone proposes Repeat each policy passage several times. Which technical objection is specifically relevant to this proposal?
- Sampling temperature does not enlarge prompt capacity
- Context windows limit how much text a model can accept
- Casing does not expand the model's token limit
- Duplication consumes capacity and can crowd out evidence
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Repeat each policy passage several times. Duplication consumes capacity and can crowd out evidence
Incorrect Answers
A:This limitation concerns Raise temperature to expand the context window; it does not diagnose the proposed Repeat each policy passage several times approach.
B:Context windows limit how much text a model can accept This explains the recommended design, but it is not the requested diagnosis of Repeat each policy passage several times.
C:This limitation concerns Use uppercase characters to increase context size; it does not diagnose the proposed Repeat each policy passage several times approach.
Question 106
An agent must access approved external APIs through a common tool interface. Which option BEST meets the business requirement?
- Use Model Context Protocol tool integration with separate authentication
- Assume mentioning a tool in a prompt makes it callable
- Use tokenization as the API authorization mechanism
- Use vector relevance as permission to transfer money
Explanation
Correct Answer
A:MCP standardizes tool connectivity but does not grant permission automatically The specified scenario makes this the most appropriate approach.
Incorrect Answers
B:Prompt text cannot establish secure tool registration This prevents the approach from meeting the stated requirement.
C:Breaking text into tokens does not authenticate API calls This prevents the approach from meeting the stated requirement.
D:Similarity is not authorization This prevents the approach from meeting the stated requirement.
Question 107
A legal assistant must draft a new paragraph from a policy topic instead of assigning a label. Before approving a rollout, an architect suggests Use clustering without a text generator. Which mismatch should be documented for this precise design?
- Grouping documents cannot compose the new paragraph
- A number does not generate requested policy wording
- A category label does not supply prose
- A generative model can compose new natural-language text
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Use clustering without a text generator. Grouping documents cannot compose the new paragraph
Incorrect Answers
B:This limitation concerns Use regression to output a monetary score as the paragraph; it does not diagnose the proposed Use clustering without a text generator approach.
C:This limitation concerns Use binary classification as the entire draft; it does not diagnose the proposed Use clustering without a text generator approach.
D:A generative model can compose new natural-language text This explains the recommended design, but it is not the requested diagnosis of Use clustering without a text generator.
Question 108
A team is selecting how to serve a production FM under predictable sustained load. Before approving a rollout, an architect suggests Train a new model on every request to guarantee throughput. Which mismatch should be documented for this precise design?
- Training per inference is not a viable serving strategy
- Capacity choice should reflect sustained traffic, latency and pricing
- On-demand services can still have limits and performance constraints
- Reserved capacity can waste money under sporadic load
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Train a new model on every request to guarantee throughput. Training per inference is not a viable serving strategy
Incorrect Answers
B:Capacity choice should reflect sustained traffic, latency and pricing This explains the recommended design, but it is not the requested diagnosis of Train a new model on every request to guarantee throughput.
C:This limitation concerns Assume on-demand never has rate or quota limits; it does not diagnose the proposed Train a new model on every request to guarantee throughput approach.
D:This limitation concerns Always buy dedicated throughput for a one-off experiment; it does not diagnose the proposed Train a new model on every request to guarantee throughput approach.
Question 109
A design studio wants new product illustrations based on descriptions. Before approving a rollout, an architect suggests Use OCR to invent the artwork. Which mismatch should be documented for this precise design?
- Sentiment does not produce synthetic images
- A numeric prediction is not the requested visual artifact
- OCR reads existing images rather than creating new ones
- Generative visual models create new imagery from prompts
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Use OCR to invent the artwork. OCR reads existing images rather than creating new ones
Incorrect Answers
A:This limitation concerns Use text sentiment labels to paint the image; it does not diagnose the proposed Use OCR to invent the artwork approach.
B:This limitation concerns Use numeric regression as the illustration generator; it does not diagnose the proposed Use OCR to invent the artwork approach.
D:Generative visual models create new imagery from prompts This explains the recommended design, but it is not the requested diagnosis of Use OCR to invent the artwork.
Question 110
An insurance assistant must interpret an uploaded photo together with a text instruction. 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 audio transcription as image understanding
- A system inventory lists this competing approach as active: Use text-only inference without a visual description
- The assistant correctly cites visible details and follows text instructions
- A reviewed design record approves this competing approach: Use a filename-based regression score as the visual interpretation
Explanation
Correct Answer
C:The assistant correctly cites visible details and follows text instructions. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Speech recognition does not process picture content A component invocation does not validate the necessary end-to-end outcome.
B:Text-only input cannot directly inspect the photo An active setting cannot independently prove the requested business outcome.
D:File names are not evidence of damage in pixels Design approval is not evidence that the alternative meets the actual requirement.
