Amazon AWS Certified AI Practitioner AIF-C01 Practice Test 27
Question 260
The initial scenario was: A legal service has reliable citations via RAG but needs a consistent tone and format across different clients. The original control is already in place. Now the application later faces stale product facts in responses. Which ADDITIONAL action BEST addresses the new condition?
- Begin with a large-scale foundation model pretraining project for tone changes
- Evaluation finds stable formatting while retrieved citations remain correct
- Improve retrieval freshness rather than relying on fine-tuning to update facts
- Compare prompt templates or light customization before costly pretraining
Explanation
Correct Answer
C:The application later faces stale product facts in responses. The appropriate added action is: Improve retrieval freshness rather than relying on fine-tuning to update facts.
Incorrect Answers
A:This original shortcut remains unsuitable: Full pretraining is disproportionate to a formatting requirement.
B:This describes evidence for the original approach, not a new action: Evaluation finds stable formatting while retrieved citations remain correct.
D:This was the original control and is already implemented. Formatting changes rarely justify training a foundation model from scratch.
Question 261
A team wants a cheaper smaller model to imitate a stronger model's useful behavior for a constrained task. Which option BEST meets the stated AI/ML or AWS requirement?
- Treat model distillation as simply deleting random training records
- Select an FM that supports both image understanding and textual output
- Rebenchmark the distilled model's capabilities before reuse
- Consider model distillation with quality and cost evaluation
Explanation
Correct Answer
D:Distillation transfers some capability from a stronger teacher to a smaller student. In this situation, choose: Consider model distillation with quality and cost evaluation.
Incorrect Answers
A:Dropping arbitrary data does not transfer teacher behavior. The proposal, Treat model distillation as simply deleting random training records, does not satisfy this need.
B:This can be valid for a different task: A retailer needs a foundation model that can interpret images of damaged parcels and answer written questions about them. The current request is different and needs the selected approach.
C:This is appropriate only after a distinct change: The task later expands to a difficult multimodal workflow. It is not the best initial response to a team wants a cheaper smaller model to imitate a stronger model's useful behavior for a constrained task.
Question 262
A team wants a cheaper smaller model to imitate a stronger model's useful behavior for a constrained task. A colleague suggests: Treat model distillation as simply deleting random training records. Which finding BEST explains why the suggestion should be rejected?
- A text-only modality cannot inspect the supplied parcel photos
- Static customization can miss current changes and does not itself provide citations
- Dropping arbitrary data does not transfer teacher behavior
- The shortcut is safe because its terminology is familiar to the team
Explanation
Correct Answer
C:Dropping arbitrary data does not transfer teacher behavior. This is the direct defect in the proposed action: Treat model distillation as simply deleting random training records.
Incorrect Answers
A:This finding is associated with a different problem: A retailer needs a foundation model that can interpret images of damaged parcels and answer written questions about them.
B:This concern belongs to another situation: A staff assistant must answer policy questions from frequently updated internal documents and display source citations.
D:Familiar wording is not proof that an action is appropriate. The actual defect is: Dropping arbitrary data does not transfer teacher behavior.
Question 263
A team wants a cheaper smaller model to imitate a stronger model's useful behavior for a constrained task. After implementing Consider model distillation with quality and cost evaluation, which observation BEST verifies the intended result?
- A pilot interprets held-out parcel photos and addresses accompanying questions
- The team schedules a future test of whether the student meets task-quality thresholds at a lower measured serving cost, but has no result
- The team approves Consider model distillation with quality and cost evaluation but does not run an operational test
- The student meets task-quality thresholds at a lower measured serving cost
Explanation
Correct Answer
D:The observed result, The student meets task-quality thresholds at a lower measured serving cost, provides direct evidence. Distillation transfers some capability from a stronger teacher to a smaller student.
Incorrect Answers
A:This observation would be evidence for another case: A retailer needs a foundation model that can interpret images of damaged parcels and answer written questions about them, not the current scenario.
B:Verify the actual outcome: The student meets task-quality thresholds at a lower measured serving cost.
C:The required observable condition is: The student meets task-quality thresholds at a lower measured serving cost.
Question 264
The initial scenario was: A team wants a cheaper smaller model to imitate a stronger model's useful behavior for a constrained task. The original control is already in place. Now the task later expands to a difficult multimodal workflow. Which ADDITIONAL action BEST addresses the new condition?
- Treat model distillation as simply deleting random training records
- Rebenchmark the distilled model's capabilities before reuse
- Consider model distillation with quality and cost evaluation
- The student meets task-quality thresholds at a lower measured serving cost
Explanation
Correct Answer
B:The task later expands to a difficult multimodal workflow. The appropriate added action is: Rebenchmark the distilled model's capabilities before reuse.
Incorrect Answers
A:This original shortcut remains unsuitable: Dropping arbitrary data does not transfer teacher behavior.
C:This was the original control and is already implemented. Distillation transfers some capability from a stronger teacher to a smaller student.
D:This describes evidence for the original approach, not a new action: The student meets task-quality thresholds at a lower measured serving cost.
Question 265
A new prompt asks a general-purpose model to categorize a support message without providing any examples. Which option BEST meets the stated AI/ML or AWS requirement?
- Identify the approach as zero-shot prompting
- Introduce a small set of representative few-shot demonstrations
- Call it few-shot learning even though no examples are shown
- Describe this as few-shot prompting with representative demonstrations
Explanation
Correct Answer
A:Zero-shot prompting supplies instructions without worked demonstrations. In this situation, choose: Identify the approach as zero-shot prompting.
Incorrect Answers
B:This is appropriate only after a distinct change: The classifier struggles with uncommon labels and reviewers supply examples. It is not the best initial response to a new prompt asks a general-purpose model to categorize a support message without providing any examples.
C:Few-shot prompting includes examples in the prompt. The proposal, Call it few-shot learning even though no examples are shown, does not satisfy this need.
D:This can be valid for a different task: A sentiment task works poorly with bare instructions, so the prompt includes three labeled example reviews. The current request is different and needs the selected approach.
Question 266
A new prompt asks a general-purpose model to categorize a support message without providing any examples. A colleague suggests: Call it few-shot learning even though no examples are shown. Which finding BEST explains why the suggestion should be rejected?
- The shortcut is safe because its terminology is familiar to the team
- Few-shot prompting includes examples in the prompt
- In-context examples condition inference without training parameter updates
- Unmanaged variations make behavior and rollback hard to reproduce
Explanation
Correct Answer
B:Few-shot prompting includes examples in the prompt. This is the direct defect in the proposed action: Call it few-shot learning even though no examples are shown.
Incorrect Answers
A:Familiar wording is not proof that an action is appropriate. The actual defect is: Few-shot prompting includes examples in the prompt.
C:This finding is associated with a different problem: A sentiment task works poorly with bare instructions, so the prompt includes three labeled example reviews.
D:This concern belongs to another situation: Several teams reuse an approved summary instruction but change the customer name and input document for each invocation.
Question 267
A new prompt asks a general-purpose model to categorize a support message without providing any examples. After implementing Identify the approach as zero-shot prompting, which observation BEST verifies the intended result?
- A test response follows the requested categories without supplied samples
- The team schedules a future test of whether a test response follows the requested categories without supplied samples, but has no result
- The team approves Identify the approach as zero-shot prompting but does not run an operational test
- New reviews match the demonstrated response convention on test cases
Explanation
Correct Answer
A:The observed result, A test response follows the requested categories without supplied samples, provides direct evidence. Zero-shot prompting supplies instructions without worked demonstrations.
Incorrect Answers
B:Verify the actual outcome: A test response follows the requested categories without supplied samples.
C:The required observable condition is: A test response follows the requested categories without supplied samples.
D:This observation would be evidence for another case: A sentiment task works poorly with bare instructions, so the prompt includes three labeled example reviews, not the current scenario.
Question 268
The initial scenario was: A new prompt asks a general-purpose model to categorize a support message without providing any examples. The original control is already in place. Now the classifier struggles with uncommon labels and reviewers supply examples. Which ADDITIONAL action BEST addresses the new condition?
- Call it few-shot learning even though no examples are shown
- A test response follows the requested categories without supplied samples
- Identify the approach as zero-shot prompting
- Introduce a small set of representative few-shot demonstrations
Explanation
Correct Answer
D:The classifier struggles with uncommon labels and reviewers supply examples. The appropriate added action is: Introduce a small set of representative few-shot demonstrations.
Incorrect Answers
A:This original shortcut remains unsuitable: Few-shot prompting includes examples in the prompt. It is not a valid solution to the changed requirement.
B:This describes evidence for the original approach, not a new action: A test response follows the requested categories without supplied samples.
C:This was the original control and is already implemented. Zero-shot prompting supplies instructions without worked demonstrations.
Question 269
A sentiment task works poorly with bare instructions, so the prompt includes three labeled example reviews. Which option BEST meets the stated AI/ML or AWS requirement?
- Describe this as few-shot prompting with representative demonstrations
- Evaluate zero-shot instruction quality for the leaner prompt
- Use a versioned prompt template with clearly named variables
- Describe labeled examples in the prompt as changing the FM's weights
Explanation
Correct Answer
A:Few-shot prompts provide examples of inputs and desired outputs. In this situation, choose: Describe this as few-shot prompting with representative demonstrations.
Incorrect Answers
B:This is appropriate only after a distinct change: The team later removes all examples to save tokens. It is not the best initial response to a sentiment task works poorly with bare instructions, so the prompt includes three labeled example reviews.
C:This can be valid for a different task: Several teams reuse an approved summary instruction but change the customer name and input document for each invocation. The current request is different and needs the selected approach.
D:In-context examples condition inference without training parameter updates. The proposal, Describe labeled examples in the prompt as changing the FM's weights, does not satisfy this need.
Question 270
A sentiment task works poorly with bare instructions, so the prompt includes three labeled example reviews. A colleague suggests: Describe labeled examples in the prompt as changing the FM's weights. Which finding BEST explains why the suggestion should be rejected?
- Open-ended formatting is unsuitable for a strict machine-readable contract
- In-context examples condition inference without training parameter updates
- Unmanaged variations make behavior and rollback hard to reproduce
- The shortcut is safe because its terminology is familiar to the team
Explanation
Correct Answer
B:In-context examples condition inference without training parameter updates. This is the direct defect in the proposed action: Describe labeled examples in the prompt as changing the FM's weights.
Incorrect Answers
A:This concern belongs to another situation: An assistant must answer in valid JSON with keys title, risk and owner for downstream parsing.
C:This finding is associated with a different problem: Several teams reuse an approved summary instruction but change the customer name and input document for each invocation.
D:Familiar wording is not proof that an action is appropriate. The actual defect is: In-context examples condition inference without training parameter updates.