AWS Certified AI Practitioner AIF-C01 Practice Test 31
Question 300
A healthcare team is curating fine-tuning examples and many contain sensitive patient identifiers. A pilot uses this approach instead: Use every raw patient record without consent assessment. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Encryption does not replace lawful data-use authorization
- Unvetted sensitive training data creates privacy and compliance risks
- Training data governance and privacy matter before model adaptation
- Weights do not by themselves guarantee training privacy
Explanation
Correct Answer
B:This is the actual failure mode of the proposed approach: Use every raw patient record without consent assessment. Unvetted sensitive training data creates privacy and compliance risks
Incorrect Answers
A:This limitation concerns Assume file encryption alone establishes the right to use all medical records; it does not diagnose the proposed Use every raw patient record without consent assessment approach.
C:Training data governance and privacy matter before model adaptation This explains the recommended design, but it is not the requested diagnosis of Use every raw patient record without consent assessment.
D:This limitation concerns Treat the model's hidden parameters as proof no patient data can leak; it does not diagnose the proposed Use every raw patient record without consent assessment approach.
Question 301
A customer service bot needs accurate answers from an internal policy collection that changes weekly. During a design review, someone proposes Increase temperature until the model recalls current internal clauses. Which technical objection is specifically relevant to this proposal?
- Static weights may not reflect weekly document changes
- Pretrained data does not contain confidential current policies by default
- Knowledge Bases can retrieve relevant documents to ground generated replies
- Randomness does not provide current policy evidence
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Increase temperature until the model recalls current internal clauses. Randomness does not provide current policy evidence
Incorrect Answers
A:This limitation concerns Fine-tune on one static snapshot as the only update mechanism; it does not diagnose the proposed Increase temperature until the model recalls current internal clauses approach.
B:This limitation concerns Rely only on public pretraining for private documents; it does not diagnose the proposed Increase temperature until the model recalls current internal clauses approach.
C:Knowledge Bases can retrieve relevant documents to ground generated replies This explains the recommended design, but it is not the requested diagnosis of Increase temperature until the model recalls current internal clauses.
Question 302
A company already runs Amazon Aurora PostgreSQL and wants vector similarity alongside existing relational records. Which option BEST meets the business requirement?
- Move all embeddings to Secrets Manager as the index
- Use the database's transaction log alone for nearest neighbors
- Assume PostgreSQL table storage without vector indexing automatically yields fast semantic search
- Evaluate the Aurora PostgreSQL vector-capable option when supported
Explanation
Correct Answer
D:A compatible relational store can combine structured data and vector retrieval The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Secret storage is not a similarity-query engine This prevents the approach from meeting the stated requirement.
B:Audit or WAL records are not a vector search index This prevents the approach from meeting the stated requirement.
C:A plain unindexed column is not equivalent to a vector search design This prevents the approach from meeting the stated requirement.
Question 303
An FM must output a compact JSON object with specified fields. During a design review, someone proposes Increase temperature to guarantee JSON validity. Which technical objection is specifically relevant to this proposal?
- Irrelevant examples do not define the output contract
- Random sampling is not a schema validator
- Concrete schema constraints and validation reduce malformed responses
- Unguided output may omit required fields
Explanation
Correct Answer
B:This is the actual failure mode of the proposed approach: Increase temperature to guarantee JSON validity. Random sampling is not a schema validator
Incorrect Answers
A:This limitation concerns Use unrelated example documents without naming desired keys; it does not diagnose the proposed Increase temperature to guarantee JSON validity approach.
C:Concrete schema constraints and validation reduce malformed responses This explains the recommended design, but it is not the requested diagnosis of Increase temperature to guarantee JSON validity.
D:This limitation concerns Provide no format guidance and rely on natural prose; it does not diagnose the proposed Increase temperature to guarantee JSON validity approach.
Question 304
An evaluation team wants scalable subjective feedback on FM answers before human sampling. Before approving a rollout, an architect suggests Use only the model's confidence in its own answer. Which mismatch should be documented for this precise design?
- Cost does not judge whether the response is correct
- Automated judges can be biased and should be calibrated
- Self-reported confidence may be poorly calibrated
- A judging model can scale rubric-based scoring but may have systematic bias
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Use only the model's confidence in its own answer. Self-reported confidence may be poorly calibrated
Incorrect Answers
A:This limitation concerns Use the price per token as the sole quality score; it does not diagnose the proposed Use only the model's confidence in its own answer approach.
B:This limitation concerns Treat judge outputs as infallible ground truth; it does not diagnose the proposed Use only the model's confidence in its own answer approach.
D:A judging model can scale rubric-based scoring but may have systematic bias This explains the recommended design, but it is not the requested diagnosis of Use only the model's confidence in its own answer.
Question 305
An assistant has a strict 300 millisecond response budget and a fixed monthly inference allowance. Which option BEST meets the business requirement?
- Benchmark smaller capable FMs for latency and unit cost before choosing
- Choose solely by advertised training dataset size
- Assume largest available model is always most responsive
- Select by price without testing task accuracy
Explanation
Correct Answer
A:A model that meets the task with less compute can be preferable The specified scenario makes this the most appropriate approach.
Incorrect Answers
B:Training corpus size does not establish served latency This prevents the approach from meeting the stated requirement.
C:Larger parameter count does not guarantee lower latency This prevents the approach from meeting the stated requirement.
D:Cheap incorrect answers fail the application requirement This prevents the approach from meeting the stated requirement.
Question 306
Two generated answers use different wording but mean much the same thing. Before approving a rollout, an architect suggests Use a classification precision statistic without a labeled classification task. Which mismatch should be documented for this precise design?
- Bytes do not capture linguistic meaning
- Semantically valid paraphrases can differ in spelling and wording
- BERTScore compares contextual representations of generated and reference text
- Precision for a separate label problem is not semantic text similarity
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Use a classification precision statistic without a labeled classification task. Precision for a separate label problem is not semantic text similarity
Incorrect Answers
A:This limitation concerns Use storage byte count as the semantic measure; it does not diagnose the proposed Use a classification precision statistic without a labeled classification task approach.
B:This limitation concerns Rely only on exact string equality; it does not diagnose the proposed Use a classification precision statistic without a labeled classification task approach.
C:BERTScore compares contextual representations of generated and reference text This explains the recommended design, but it is not the requested diagnosis of Use a classification precision statistic without a labeled classification task.
Question 307
A retailer needs a recurring branded tone across thousands of responses based on a stable style guide. During a design review, someone proposes Use a token quota increase as the style training method. Which technical objection is specifically relevant to this proposal?
- Quota changes do not teach the desired tone
- Weight adaptation can improve consistent behavior for a stable specialized style
- Similarity search alone is not a trained response style
- Retrieval supplies context but does not persistently alter weights
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Use a token quota increase as the style training method. Quota changes do not teach the desired tone
Incorrect Answers
B:Weight adaptation can improve consistent behavior for a stable specialized style This explains the recommended design, but it is not the requested diagnosis of Use a token quota increase as the style training method.
C:This limitation concerns Rely only on an embedding vector index to enforce every phrasing choice; it does not diagnose the proposed Use a token quota increase as the style training method approach.
D:This limitation concerns Assume RAG always changes the model's learned style permanently; it does not diagnose the proposed Use a token quota increase as the style training method approach.
Question 308
A document summarizer changed prompt wording and stakeholders need to compare the revision objectively. 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: Judge only by how long the prompt is
- A system inventory lists this competing approach as active: Compare unrelated documents before and after the edit
- A reviewed design record approves this competing approach: Assume a new prompt name proves improved output
- Two prompt versions are scored on identical reviewed inputs and documented criteria
Explanation
Correct Answer
D:Two prompt versions are scored on identical reviewed inputs and documented criteria. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Prompt length alone does not measure answer quality A component invocation does not validate the necessary end-to-end outcome.
B:Different inputs confound prompt-version comparisons An active setting cannot independently prove the requested business outcome.
C:Renaming does not test relevance or accuracy Design approval is not evidence that the alternative meets the actual requirement.
Question 309
An internal prompt includes confidential API keys that users might echo back. Before approving a rollout, an architect suggests Assume model refusal alone makes a leaked key invalid. Which mismatch should be documented for this precise design?
- Obscurity is not credential protection
- Prompts may be logged or exposed and should not contain embedded credentials
- Refusal behavior cannot replace credential rotation and least privilege
- Sampling randomness does not revoke exposed credentials
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Assume model refusal alone makes a leaked key invalid. Refusal behavior cannot replace credential rotation and least privilege
Incorrect Answers
A:This limitation concerns Rely solely on hiding secret text in a longer prompt; it does not diagnose the proposed Assume model refusal alone makes a leaked key invalid approach.
B:Prompts may be logged or exposed and should not contain embedded credentials This explains the recommended design, but it is not the requested diagnosis of Assume model refusal alone makes a leaked key invalid.
D:This limitation concerns Raise temperature so the model forgets the secret; it does not diagnose the proposed Assume model refusal alone makes a leaked key invalid approach.
Question 310
A specialized science assistant requires substantial domain-language familiarity absent from generic pretraining. A pilot uses this approach instead: Use prompt caching as the only way to teach unfamiliar specialized language. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Caching reused prefixes does not train model weights
- Presentation changes do not alter model language knowledge
- RAG helps facts but may not substitute for deep domain adaptation
- Further training on domain corpora can adapt foundational representations
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Use prompt caching as the only way to teach unfamiliar specialized language. Caching reused prefixes does not train model weights
Incorrect Answers
B:This limitation concerns Treat a new user interface as domain adaptation; it does not diagnose the proposed Use prompt caching as the only way to teach unfamiliar specialized language approach.
C:This limitation concerns Assume retrieval alone always yields the desired base domain language behavior; it does not diagnose the proposed Use prompt caching as the only way to teach unfamiliar specialized language approach.
D:Further training on domain corpora can adapt foundational representations This explains the recommended design, but it is not the requested diagnosis of Use prompt caching as the only way to teach unfamiliar specialized language.
