Amazon AWS Certified AI Practitioner AIF-C01 Practice Test 35
Question 340
The initial scenario was: A localization group evaluates short machine translations against approved human reference translations. The original control is already in place. Now authors later produce paraphrases with different wording but similar meaning. Which ADDITIONAL action BEST addresses the new condition?
- Add semantic measures or human adequacy judgments
- Evaluate BLEU as one reference-based translation metric
- Use server uptime percentage as the sole language quality metric
- BLEU and human checks are reported for a representative translation test set
Explanation
Correct Answer
A:Authors later produce paraphrases with different wording but similar meaning. The appropriate added action is: Add semantic measures or human adequacy judgments.
Incorrect Answers
B:This was the original control and is already implemented. BLEU compares n-gram overlap and is often used for translation benchmarking.
C:This original shortcut remains unsuitable: Availability does not measure semantic quality of translations.
D:This describes evidence for the original approach, not a new action: BLEU and human checks are reported for a representative translation test set.
Question 341
A company finds simple word overlap scores underrate accurate paraphrased text in a generated answer. Which option BEST meets the stated AI/ML or AWS requirement?
- Demand exact token identity as the only correctness criterion
- Consider BERTScore or semantic comparison alongside other checks
- Add factuality checks against authoritative evidence and reviewer assessments
- Use a repeatable model evaluation protocol, including Amazon Bedrock Model Evaluation where suitable
Explanation
Correct Answer
B:Embedding-based comparison can capture semantic similarity beyond literal word overlap. In this situation, choose: Consider BERTScore or semantic comparison alongside other checks.
Incorrect Answers
A:Equivalent meanings can be expressed with different word sequences. The proposal, Demand exact token identity as the only correctness criterion, does not satisfy this need.
C:This is appropriate only after a distinct change: The team later needs assurance that cited facts are actually true. It is not the best initial response to a company finds simple word overlap scores underrate accurate paraphrased text in a generated answer.
D:This can be valid for a different task: A company evaluates multiple Bedrock foundation models using the same task dataset and human quality judgments. The current request is different and needs the selected approach.
Question 342
A company finds simple word overlap scores underrate accurate paraphrased text in a generated answer. A colleague suggests: Demand exact token identity as the only correctness criterion. Which finding BEST explains why the suggestion should be rejected?
- Text similarity alone does not guarantee clinical truth or safe advice
- Equivalent meanings can be expressed with different word sequences
- The shortcut is safe because its terminology is familiar to the team
- Brand familiarity is not a model quality benchmark
Explanation
Correct Answer
B:Equivalent meanings can be expressed with different word sequences. This is the direct defect in the proposed action: Demand exact token identity as the only correctness criterion.
Incorrect Answers
A:This concern belongs to another situation: A medical assistant may be fluent but wrong in uncommon urgent scenarios, requiring reviewers to inspect answers.
C:Familiar wording is not proof that an action is appropriate. The actual defect is: Equivalent meanings can be expressed with different word sequences.
D:This finding is associated with a different problem: A company evaluates multiple Bedrock foundation models using the same task dataset and human quality judgments.
Question 343
A company finds simple word overlap scores underrate accurate paraphrased text in a generated answer. After implementing Consider BERTScore or semantic comparison alongside other checks, which observation BEST verifies the intended result?
- A validation set shows semantically aligned paraphrases rated more reasonably
- The team approves Consider BERTScore or semantic comparison alongside other checks but does not run an operational test
- The team schedules a future test of whether a validation set shows semantically aligned paraphrases rated more reasonably, but has no result
- Evaluations report consistent task metrics and selected human ratings
Explanation
Correct Answer
A:The observed result, A validation set shows semantically aligned paraphrases rated more reasonably, provides direct evidence. Embedding-based comparison can capture semantic similarity beyond literal word overlap.
Incorrect Answers
B:The required observable condition is: A validation set shows semantically aligned paraphrases rated more reasonably.
C:Verify the actual outcome: A validation set shows semantically aligned paraphrases rated more reasonably.
D:This observation would be evidence for another case: A company evaluates multiple Bedrock foundation models using the same task dataset and human quality judgments, not the current scenario.
Question 344
The initial scenario was: A company finds simple word overlap scores underrate accurate paraphrased text in a generated answer. The original control is already in place. Now the team later needs assurance that cited facts are actually true. Which ADDITIONAL action BEST addresses the new condition?
- Consider BERTScore or semantic comparison alongside other checks
- Demand exact token identity as the only correctness criterion
- Add factuality checks against authoritative evidence and reviewer assessments
- A validation set shows semantically aligned paraphrases rated more reasonably
Explanation
Correct Answer
C:The team later needs assurance that cited facts are actually true. The appropriate added action is: Add factuality checks against authoritative evidence and reviewer assessments.
Incorrect Answers
A:This was the original control and is already implemented. Embedding-based comparison can capture semantic similarity beyond literal word overlap.
B:This original shortcut remains unsuitable: Equivalent meanings can be expressed with different word sequences.
D:This describes evidence for the original approach, not a new action: A validation set shows semantically aligned paraphrases rated more reasonably.
Question 345
A company evaluates multiple Bedrock foundation models using the same task dataset and human quality judgments. Which option BEST meets the stated AI/ML or AWS requirement?
- Use a repeatable model evaluation protocol, including Amazon Bedrock Model Evaluation where suitable
- Choose the model with the most attractive name without testing
- Add load and response-time benchmarks to the evaluation
- Use human-in-the-loop evaluation for high-impact sample decisions
Explanation
Correct Answer
A:Comparable prompts and metrics support evidence-based model selection. In this situation, choose: Use a repeatable model evaluation protocol, including Amazon Bedrock Model Evaluation where suitable.
Incorrect Answers
B:Brand familiarity is not a model quality benchmark. The proposal, Choose the model with the most attractive name without testing, does not satisfy this need.
C:This is appropriate only after a distinct change: A new production task introduces additional latency constraints. It is not the best initial response to a company evaluates multiple bedrock foundation models using the same task dataset and human quality judgments.
D:This can be valid for a different task: A medical assistant may be fluent but wrong in uncommon urgent scenarios, requiring reviewers to inspect answers. The current request is different and needs the selected approach.
Question 346
A company evaluates multiple Bedrock foundation models using the same task dataset and human quality judgments. A colleague suggests: Choose the model with the most attractive name without testing. Which finding BEST explains why the suggestion should be rejected?
- Text generation parameters do not fix missing source passages
- The shortcut is safe because its terminology is familiar to the team
- Text similarity alone does not guarantee clinical truth or safe advice
- Brand familiarity is not a model quality benchmark
Explanation
Correct Answer
D:Brand familiarity is not a model quality benchmark. This is the direct defect in the proposed action: Choose the model with the most attractive name without testing.
Incorrect Answers
A:This concern belongs to another situation: A RAG assistant produces incorrect answers because the relevant document often fails to appear among retrieved chunks.
B:Familiar wording is not proof that an action is appropriate. The actual defect is: Brand familiarity is not a model quality benchmark.
C:This finding is associated with a different problem: A medical assistant may be fluent but wrong in uncommon urgent scenarios, requiring reviewers to inspect answers.
Question 347
A company evaluates multiple Bedrock foundation models using the same task dataset and human quality judgments. After implementing Use a repeatable model evaluation protocol, including Amazon Bedrock Model Evaluation where suitable, which observation BEST verifies the intended result?
- Evaluations report consistent task metrics and selected human ratings
- The team approves Use a repeatable model evaluation protocol, including Amazon Bedrock Model Evaluation where suitable but does not run an operational test
- Qualified reviewers assess sampled high-risk outputs and record corrections
- The team schedules a future test of whether evaluations report consistent task metrics and selected human ratings, but has no result
Explanation
Correct Answer
A:The observed result, Evaluations report consistent task metrics and selected human ratings, provides direct evidence. Comparable prompts and metrics support evidence-based model selection.
Incorrect Answers
B:The required observable condition is: Evaluations report consistent task metrics and selected human ratings.
C:This observation would be evidence for another case: A medical assistant may be fluent but wrong in uncommon urgent scenarios, requiring reviewers to inspect answers, not the current scenario.
D:An unperformed future check is not evidence. Verify the actual outcome: Evaluations report consistent task metrics and selected human ratings.
Question 348
The initial scenario was: A company evaluates multiple Bedrock foundation models using the same task dataset and human quality judgments. The original control is already in place. Now a new production task introduces additional latency constraints. Which ADDITIONAL action BEST addresses the new condition?
- Choose the model with the most attractive name without testing
- Evaluations report consistent task metrics and selected human ratings
- Add load and response-time benchmarks to the evaluation
- Use a repeatable model evaluation protocol, including Amazon Bedrock Model Evaluation where suitable
Explanation
Correct Answer
C:A new production task introduces additional latency constraints. The appropriate added action is: Add load and response-time benchmarks to the evaluation.
Incorrect Answers
A:This original shortcut remains unsuitable: Brand familiarity is not a model quality benchmark. It is not a valid solution to the changed requirement.
B:This describes evidence for the original approach, not a new action: Evaluations report consistent task metrics and selected human ratings.
D:This was the original control and is already implemented. Comparable prompts and metrics support evidence-based model selection.
Question 349
A medical assistant may be fluent but wrong in uncommon urgent scenarios, requiring reviewers to inspect answers. Which TWO statements jointly identify the best initial approach and an observable verification result? (Select TWO.)
- Expand review to multilingual and subgroup samples
- Evaluate retrieval relevance and recall-at-k separately from generation quality
- Treat an automated similarity score as proof all responses are medically correct
- Use human-in-the-loop evaluation for high-impact sample decisions
- Qualified reviewers assess sampled high-risk outputs and record corrections
Explanation
Correct Answers
D:Correct approach. Expert human review can identify harmful or factually wrong outputs that automatic scores miss. This directly serves the stated situation: A medical assistant may be fluent but wrong in uncommon urgent scenarios, requiring reviewers to inspect answers.
E:Correct operating evidence. Observing qualified reviewers assess sampled high-risk outputs and record corrections verifies this approach instead of relying on unperformed plans.
Incorrect Answers
A:This would be appropriate if the separate requirement arose: The service adds more languages used by patients. That is not the present condition.
B:This addresses a different scenario: A RAG assistant produces incorrect answers because the relevant document often fails to appear among retrieved chunks. It does not constitute the requested approach or operating result for the current case.
C:Text similarity alone does not guarantee clinical truth or safe advice. The suggested shortcut would not meet the present requirement.
Question 350
A medical assistant may be fluent but wrong in uncommon urgent scenarios, requiring reviewers to inspect answers. A colleague suggests: Treat an automated similarity score as proof all responses are medically correct. Which finding BEST explains why the suggestion should be rejected?
- The shortcut is safe because its terminology is familiar to the team
- Text generation parameters do not fix missing source passages
- Text similarity alone does not guarantee clinical truth or safe advice
- More tool calls may indicate waste or unsafe actions, not success
Explanation
Correct Answer
C:Text similarity alone does not guarantee clinical truth or safe advice. This is the direct defect in the proposed action: Treat an automated similarity score as proof all responses are medically correct.
Incorrect Answers
A:Familiar wording is not proof that an action is appropriate. The actual defect is: Text similarity alone does not guarantee clinical truth or safe advice.
B:This finding is associated with a different problem: A RAG assistant produces incorrect answers because the relevant document often fails to appear among retrieved chunks.
D:This concern belongs to another situation: A tool-using agent finishes some tasks correctly but occasionally performs unnecessary API calls.