Amazon AWS Certified AI Practitioner AIF-C01 Practice Test 16
Question 150
An insurance workflow uses one agent to collect forms and another to verify policy coverage before final review. A colleague suggests: Let all agents write unrestricted final decisions without coordination. Which finding BEST explains why the suggestion should be rejected?
- The shortcut is safe because its terminology is familiar to the team
- A conversation can contain many tokens across its context and generated text
- Giant documents are difficult to search and may exceed model context limits
- Uncontrolled agents lack traceable assignment and review boundaries
Explanation
Correct Answer
D:Uncontrolled agents lack traceable assignment and review boundaries. This is the direct defect in the proposed action: Let all agents write unrestricted final decisions without coordination.
Incorrect Answers
A:Familiar wording is not proof that an action is appropriate. The actual defect is: Uncontrolled agents lack traceable assignment and review boundaries.
B:This finding is associated with a different problem: A support assistant's usage report bills for input and generated output even though the request contains no images.
C:This concern belongs to another situation: A team indexes 600-page manuals so retrieval finds relevant paragraphs rather than sending every page into a model.
Question 151
An insurance workflow uses one agent to collect forms and another to verify policy coverage before final review. After implementing Use orchestrated multi-agent collaboration with explicit handoffs, which observation BEST verifies the intended result?
- Monitoring shows input and output token totals on representative requests
- Logs show each specialist's output and a governed handoff to the reviewer
- The team schedules a future test of whether logs show each specialist's output and a governed handoff to the reviewer, but has no result
- The team approves Use orchestrated multi-agent collaboration with explicit handoffs but does not run an operational test
Explanation
Correct Answer
B:The observed result, Logs show each specialist's output and a governed handoff to the reviewer, provides direct evidence. Specialized agents can divide work while passing context and results under controls.
Incorrect Answers
A:This observation would be evidence for another case: A support assistant's usage report bills for input and generated output even though the request contains no images, not the current scenario.
C:Verify the actual outcome: Logs show each specialist's output and a governed handoff to the reviewer.
D:The required observable condition is: Logs show each specialist's output and a governed handoff to the reviewer.
Question 152
The initial scenario was: An insurance workflow uses one agent to collect forms and another to verify policy coverage before final review. The original control is already in place. Now a critical decision must always receive a human approval. Which ADDITIONAL action BEST addresses the new condition?
- Let all agents write unrestricted final decisions without coordination
- Introduce human-in-the-loop approval before the final action
- Logs show each specialist's output and a governed handoff to the reviewer
- Use orchestrated multi-agent collaboration with explicit handoffs
Explanation
Correct Answer
B:A critical decision must always receive a human approval. The appropriate added action is: Introduce human-in-the-loop approval before the final action.
Incorrect Answers
A:This original shortcut remains unsuitable: Uncontrolled agents lack traceable assignment and review boundaries.
C:This describes evidence for the original approach, not a new action: Logs show each specialist's output and a governed handoff to the reviewer.
D:This was the original control and is already implemented. Specialized agents can divide work while passing context and results under controls.
Question 153
A legal-help chatbot sometimes invents court citations that cannot be located in the source material. Which option BEST meets the stated AI/ML or AWS requirement?
- Treat unsupported citations as hallucinations and require evidence checking
- Ground answers through retrieval with validated source links
- Explain nondeterministic generation and evaluate answer consistency
- Accept fluent citations as verified solely because they look formal
Explanation
Correct Answer
A:Generative models can produce plausible but false factual statements. In this situation, choose: Treat unsupported citations as hallucinations and require evidence checking.
Incorrect Answers
B:This is appropriate only after a distinct change: The team introduces an internal searchable case repository. It is not the best initial response to a legal-help chatbot sometimes invents court citations that cannot be located in the source material.
C:This can be valid for a different task: A product manager reruns the same open-ended prompt and receives different wording even when no data changed. The current request is different and needs the selected approach.
D:Convincing wording does not demonstrate source authenticity. The proposal, Accept fluent citations as verified solely because they look formal, does not satisfy this need.
Question 154
A legal-help chatbot sometimes invents court citations that cannot be located in the source material. A colleague suggests: Accept fluent citations as verified solely because they look formal. Which finding BEST explains why the suggestion should be rejected?
- The shortcut is safe because its terminology is familiar to the team
- Model size alone does not define business value or cost-efficiency
- Generated model responses can differ without a changed request
- Convincing wording does not demonstrate source authenticity
Explanation
Correct Answer
D:Convincing wording does not demonstrate source authenticity. This is the direct defect in the proposed action: Accept fluent citations as verified solely because they look formal.
Incorrect Answers
A:Familiar wording is not proof that an action is appropriate. The actual defect is: Convincing wording does not demonstrate source authenticity.
B:This concern belongs to another situation: A business considers a very large model even though a smaller model already meets its classification and latency requirements.
C:This finding is associated with a different problem: A product manager reruns the same open-ended prompt and receives different wording even when no data changed.
Question 155
A legal-help chatbot sometimes invents court citations that cannot be located in the source material. After implementing Treat unsupported citations as hallucinations and require evidence checking, which observation BEST verifies the intended result?
- The team schedules a future test of whether reviewers can trace each cited legal statement to an actual authorized source, but has no result
- Quality measurements compare multiple runs for semantic correctness
- The team approves Treat unsupported citations as hallucinations and require evidence checking but does not run an operational test
- Reviewers can trace each cited legal statement to an actual authorized source
Explanation
Correct Answer
D:The observed result, Reviewers can trace each cited legal statement to an actual authorized source, provides direct evidence. Generative models can produce plausible but false factual statements.
Incorrect Answers
A:Verify the actual outcome: Reviewers can trace each cited legal statement to an actual authorized source.
B:This observation would be evidence for another case: A product manager reruns the same open-ended prompt and receives different wording even when no data changed, not the current scenario.
C:The required observable condition is: Reviewers can trace each cited legal statement to an actual authorized source.
Question 156
The initial scenario was: A legal-help chatbot sometimes invents court citations that cannot be located in the source material. The original control is already in place. Now the team introduces an internal searchable case repository. Which ADDITIONAL action BEST addresses the new condition?
- Accept fluent citations as verified solely because they look formal
- Treat unsupported citations as hallucinations and require evidence checking
- Ground answers through retrieval with validated source links
- Reviewers can trace each cited legal statement to an actual authorized source
Explanation
Correct Answer
C:The team introduces an internal searchable case repository. The appropriate added action is: Ground answers through retrieval with validated source links.
Incorrect Answers
A:This original shortcut remains unsuitable: Convincing wording does not demonstrate source authenticity.
B:This was the original control and is already implemented. Generative models can produce plausible but false factual statements.
D:This describes evidence for the original approach, not a new action: Reviewers can trace each cited legal statement to an actual authorized source.
Question 157
A product manager reruns the same open-ended prompt and receives different wording even when no data changed. Which TWO statements jointly identify the best initial approach and an observable verification result? (Select TWO.)
- Explain nondeterministic generation and evaluate answer consistency
- Compare quality, response time and cost before selecting model size
- Quality measurements compare multiple runs for semantic correctness
- Assume every FM invocation must return identical character-for-character text
- Serve approved deterministic template text for the fixed passage
Explanation
Correct Answers
A:Correct approach. Sampling and model behavior can vary outputs between invocations. This directly serves the stated situation: A product manager reruns the same open-ended prompt and receives different wording even when no data changed.
C:Correct operating evidence. Observing quality measurements compare multiple runs for semantic correctness verifies this approach instead of relying on unperformed plans.
Incorrect Answers
B:This addresses a different scenario: A business considers a very large model even though a smaller model already meets its classification and latency requirements. It does not constitute the requested approach or operating result for the current case.
D:Generated model responses can differ without a changed request. The suggested shortcut would not meet the present requirement.
E:This would be appropriate if the separate requirement arose: Compliance later requires exact legal boilerplate every time. That is not the present condition.
Question 158
A product manager reruns the same open-ended prompt and receives different wording even when no data changed. A colleague suggests: Assume every FM invocation must return identical character-for-character text. Which finding BEST explains why the suggestion should be rejected?
- More tokens do not directly show successful request resolution
- Generated model responses can differ without a changed request
- Model size alone does not define business value or cost-efficiency
- The shortcut is safe because its terminology is familiar to the team
Explanation
Correct Answer
B:Generated model responses can differ without a changed request. This is the direct defect in the proposed action: Assume every FM invocation must return identical character-for-character text.
Incorrect Answers
A:This concern belongs to another situation: A startup pilots a FAQ assistant and wants measurable business value rather than counting generated paragraphs.
C:This finding is associated with a different problem: A business considers a very large model even though a smaller model already meets its classification and latency requirements.
D:Familiar wording is not proof that an action is appropriate. The actual defect is: Generated model responses can differ without a changed request.
Question 159
A product manager reruns the same open-ended prompt and receives different wording even when no data changed. After implementing Explain nondeterministic generation and evaluate answer consistency, which observation BEST verifies the intended result?
- The team schedules a future test of whether quality measurements compare multiple runs for semantic correctness, but has no result
- A benchmark shows the chosen model meets quality and latency at target unit cost
- The team approves Explain nondeterministic generation and evaluate answer consistency but does not run an operational test
- Quality measurements compare multiple runs for semantic correctness
Explanation
Correct Answer
D:The observed result, Quality measurements compare multiple runs for semantic correctness, provides direct evidence. Sampling and model behavior can vary outputs between invocations.
Incorrect Answers
A:An unperformed future check is not evidence. Verify the actual outcome: Quality measurements compare multiple runs for semantic correctness.
B:This observation would be evidence for another case: A business considers a very large model even though a smaller model already meets its classification and latency requirements, not the current scenario.
C:The required observable condition is: Quality measurements compare multiple runs for semantic correctness.
Question 160
The initial scenario was: A product manager reruns the same open-ended prompt and receives different wording even when no data changed. The original control is already in place. Now compliance later requires exact legal boilerplate every time. Which ADDITIONAL action BEST addresses the new condition?
- Assume every FM invocation must return identical character-for-character text
- Serve approved deterministic template text for the fixed passage
- Explain nondeterministic generation and evaluate answer consistency
- Quality measurements compare multiple runs for semantic correctness
Explanation
Correct Answer
B:Compliance later requires exact legal boilerplate every time. The appropriate added action is: Serve approved deterministic template text for the fixed passage.
Incorrect Answers
A:This original shortcut remains unsuitable: Generated model responses can differ without a changed request.
C:This was the original control and is already implemented. Sampling and model behavior can vary outputs between invocations.
D:This describes evidence for the original approach, not a new action: Quality measurements compare multiple runs for semantic correctness.