AWS Certified AI Practitioner AIF-C01 Practice Test 32
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.
Question 311
A contract model must infer the desired classification format from several labeled examples supplied in a prompt. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- Held-out inputs follow the illustrated classification format
- A reviewed design record approves this competing approach: Replace examples with a larger token billing limit
- A system inventory lists this competing approach as active: Call a prompt with no examples few-shot
- A component test invokes this alternate approach successfully: Use a random list of unrelated paragraphs as examples
Explanation
Correct Answer
A:Held-out inputs follow the illustrated classification format. This checks the business result rather than just a configured feature.
Incorrect Answers
B:Billing settings do not provide task examples Design approval is not evidence that the alternative meets the actual requirement.
C:No demonstrations makes it zero-shot An active setting cannot independently prove the requested business outcome.
D:Unrelated demonstrations may confuse the mapping A component invocation does not validate the necessary end-to-end outcome.
Question 312
Two generated answers use different wording but mean much the same thing. 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 storage byte count as the semantic measure
- A reviewed design record approves this competing approach: Use a classification precision statistic without a labeled classification task
- Human-accepted paraphrases receive appropriate semantic scores in validation
- A system inventory lists this competing approach as active: Rely only on exact string equality
Explanation
Correct Answer
C:Human-accepted paraphrases receive appropriate semantic scores in validation. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Bytes do not capture linguistic meaning A component invocation does not validate the necessary end-to-end outcome.
B:Precision for a separate label problem is not semantic text similarity Design approval is not evidence that the alternative meets the actual requirement.
D:Semantically valid paraphrases can differ in spelling and wording An active setting cannot independently prove the requested business outcome.
Question 313
A document search app needs managed search across embedding vectors generated from product descriptions. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- Paraphrased test queries retrieve relevant products with measurable recall
- A system inventory lists this competing approach as active: Put only prose in CloudTrail audit logs for semantic lookup
- A reviewed design record approves this competing approach: Use AWS Config compliance timelines as a vector store
- A component test invokes this alternate approach successfully: Use KMS keys as the embedding database
Explanation
Correct Answer
A:Paraphrased test queries retrieve relevant products with measurable recall. This checks the business result rather than just a configured feature.
Incorrect Answers
B:Audit logs are not designed as the search similarity index An active setting cannot independently prove the requested business outcome.
C:Resource-configuration history is not a vector similarity engine Design approval is not evidence that the alternative meets the actual requirement.
D:Encryption key storage cannot evaluate nearest-neighbor queries A component invocation does not validate the necessary end-to-end outcome.
Question 314
A summarization model must capture key ideas from reference summaries. Before approving a rollout, an architect suggests Treat low latency alone as summary faithfulness. Which mismatch should be documented for this precise design?
- ROUGE compares overlap with reference text but misses some semantic quality
- Key configuration does not evaluate language output
- Fast output may still omit or invent facts
- Training loss does not directly measure summary quality against a reference
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Treat low latency alone as summary faithfulness. Fast output may still omit or invent facts
Incorrect Answers
A:ROUGE compares overlap with reference text but misses some semantic quality This explains the recommended design, but it is not the requested diagnosis of Treat low latency alone as summary faithfulness.
B:This limitation concerns Use AWS KMS encryption status as a summary metric; it does not diagnose the proposed Treat low latency alone as summary faithfulness approach.
D:This limitation concerns Use only training loss to score final user-facing summaries; it does not diagnose the proposed Treat low latency alone as summary faithfulness approach.
Question 315
A firm wants to adapt a pretrained FM to sector-specific instructions without training it from scratch. A pilot uses this approach instead: Pretrain an entirely new large model for a small style adjustment. Which limitation MOST DIRECTLY explains why the requirement is not met?
- A quota increase is not a model adaptation run
- Full pretraining demands much greater data and compute
- Fine-tuning modifies learned behavior using a smaller specialized dataset
- Routing infrastructure does not fine-tune model parameters
Explanation
Correct Answer
B:This is the actual failure mode of the proposed approach: Pretrain an entirely new large model for a small style adjustment. Full pretraining demands much greater data and compute
Incorrect Answers
A:This limitation concerns Only increase prompt token allowance as the training step; it does not diagnose the proposed Pretrain an entirely new large model for a small style adjustment approach.
C:Fine-tuning modifies learned behavior using a smaller specialized dataset This explains the recommended design, but it is not the requested diagnosis of Pretrain an entirely new large model for a small style adjustment.
D:This limitation concerns Assume adding a new API Gateway route changes the model weights; it does not diagnose the proposed Pretrain an entirely new large model for a small style adjustment approach.
Question 316
An HR assistant's answers need recently revised procedures without changing the base model's style. 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: Store updated files but never retrieve them during answering
- The next question references the revised procedure and supporting citation
- A system inventory lists this competing approach as active: Retrain the entire foundation model on every HR policy edit
- A component test invokes this alternate approach successfully: Change temperature and assume it refreshes factual weights
Explanation
Correct Answer
B:The next question references the revised procedure and supporting citation. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Unused documents cannot ground generation Design approval is not evidence that the alternative meets the actual requirement.
C:Full pretraining for frequent policy updates is excessive An active setting cannot independently prove the requested business outcome.
D:Sampling changes cannot load new documents A component invocation does not validate the necessary end-to-end outcome.
Question 317
A translation model's output must be compared to reference translations. Before approving a rollout, an architect suggests Treat request throughput as semantic translation accuracy. Which mismatch should be documented for this precise design?
- Speed is not fidelity of meaning
- BLEU evaluates n-gram overlap for candidate and reference translations
- Binary classification counts do not compare translated sentences
- Size does not establish reference translation match
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Treat request throughput as semantic translation accuracy. Speed is not fidelity of meaning
Incorrect Answers
B:BLEU evaluates n-gram overlap for candidate and reference translations This explains the recommended design, but it is not the requested diagnosis of Treat request throughput as semantic translation accuracy.
C:This limitation concerns Use a fraud confusion matrix as the translation-quality metric; it does not diagnose the proposed Treat request throughput as semantic translation accuracy approach.
D:This limitation concerns Use only model parameter count to score translations; it does not diagnose the proposed Treat request throughput as semantic translation accuracy approach.
Question 318
A feedback program collects human preference pairs for generated customer responses. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- A system inventory lists this competing approach as active: Use unlabeled random characters as preference rankings
- Independent raters prefer adapted responses on the defined policy and helpfulness criteria
- A reviewed design record approves this competing approach: Replace preference review with only storage encryption
- A component test invokes this alternate approach successfully: Assume a token-count report is an alignment label
Explanation
Correct Answer
B:Independent raters prefer adapted responses on the defined policy and helpfulness criteria. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Uninterpretable data cannot provide reliable preference signals An active setting cannot independently prove the requested business outcome.
C:Encryption protects data but does not teach response preferences Design approval is not evidence that the alternative meets the actual requirement.
D:Token volume does not express which answer humans prefer A component invocation does not validate the necessary end-to-end outcome.
Question 319
An HR assistant's answers need recently revised procedures without changing the base model's style. Which option BEST meets the business requirement?
- Prefer RAG with regularly refreshed source documents
- Retrain the entire foundation model on every HR policy edit
- Store updated files but never retrieve them during answering
- Change temperature and assume it refreshes factual weights
Explanation
Correct Answer
A:Retrieval updates facts without requiring repeated model training The specified scenario makes this the most appropriate approach.
Incorrect Answers
B:Full pretraining for frequent policy updates is excessive This prevents the approach from meeting the stated requirement.
C:Unused documents cannot ground generation This prevents the approach from meeting the stated requirement.
D:Sampling changes cannot load new documents This prevents the approach from meeting the stated requirement.
Question 320
A team wants to train a smaller assistant using responses from a larger teacher model. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- The trained student meets task benchmarks with lower serving cost
- A reviewed design record approves this competing approach: Give the same large teacher every request with no student
- A component test invokes this alternate approach successfully: Only compress prompt text and claim the model was distilled
- A system inventory lists this competing approach as active: Use a vector index as the student weights
Explanation
Correct Answer
A:The trained student meets task benchmarks with lower serving cost. This checks the business result rather than just a configured feature.
Incorrect Answers
B:Serving the teacher alone does not create a smaller trained model Design approval is not evidence that the alternative meets the actual requirement.
C:Prompt compression does not transfer learned parameters A component invocation does not validate the necessary end-to-end outcome.
D:Retrieval vectors do not replace student training An active setting cannot independently prove the requested business outcome.
