AWS Certified AI Practitioner AIF-C01 Practice Test 25
Question 240
A summarization model must capture key ideas from reference summaries. Which option BEST meets the business requirement?
- Use only training loss to score final user-facing summaries
- Use ROUGE as one reference-overlap evaluation signal, complemented by human review
- Treat low latency alone as summary faithfulness
- Use AWS KMS encryption status as a summary metric
Explanation
Correct Answer
B:ROUGE compares overlap with reference text but misses some semantic quality The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Training loss does not directly measure summary quality against a reference This prevents the approach from meeting the stated requirement.
C:Fast output may still omit or invent facts This prevents the approach from meeting the stated requirement.
D:Key configuration does not evaluate language output This prevents the approach from meeting the stated requirement.
Question 241
A company already runs Amazon Aurora PostgreSQL and wants vector similarity alongside existing relational records. Before approving a rollout, an architect suggests Assume PostgreSQL table storage without vector indexing automatically yields fast semantic search. Which mismatch should be documented for this precise design?
- A plain unindexed column is not equivalent to a vector search design
- Audit or WAL records are not a vector search index
- Secret storage is not a similarity-query engine
- A compatible relational store can combine structured data and vector retrieval
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Assume PostgreSQL table storage without vector indexing automatically yields fast semantic search. A plain unindexed column is not equivalent to a vector search design
Incorrect Answers
B:This limitation concerns Use the database's transaction log alone for nearest neighbors; it does not diagnose the proposed Assume PostgreSQL table storage without vector indexing automatically yields fast semantic search approach.
C:This limitation concerns Move all embeddings to Secrets Manager as the index; it does not diagnose the proposed Assume PostgreSQL table storage without vector indexing automatically yields fast semantic search approach.
D:A compatible relational store can combine structured data and vector retrieval This explains the recommended design, but it is not the requested diagnosis of Assume PostgreSQL table storage without vector indexing automatically yields fast semantic search.
Question 242
An HR assistant's answers need recently revised procedures without changing the base model's style. During a design review, someone proposes Change temperature and assume it refreshes factual weights. Which technical objection is specifically relevant to this proposal?
- Unused documents cannot ground generation
- Full pretraining for frequent policy updates is excessive
- Retrieval updates facts without requiring repeated model training
- Sampling changes cannot load new documents
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Change temperature and assume it refreshes factual weights. Sampling changes cannot load new documents
Incorrect Answers
A:This limitation concerns Store updated files but never retrieve them during answering; it does not diagnose the proposed Change temperature and assume it refreshes factual weights approach.
B:This limitation concerns Retrain the entire foundation model on every HR policy edit; it does not diagnose the proposed Change temperature and assume it refreshes factual weights approach.
C:Retrieval updates facts without requiring repeated model training This explains the recommended design, but it is not the requested diagnosis of Change temperature and assume it refreshes factual weights.
Question 243
A retailer needs a recurring branded tone across thousands of responses based on a stable style guide. 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 a token quota increase as the style training method
- Held-out customer prompts follow the brand tone without factual degradation
- A system inventory lists this competing approach as active: Assume RAG always changes the model's learned style permanently
- A reviewed design record approves this competing approach: Rely only on an embedding vector index to enforce every phrasing choice
Explanation
Correct Answer
B:Held-out customer prompts follow the brand tone without factual degradation. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Quota changes do not teach the desired tone A component invocation does not validate the necessary end-to-end outcome.
C:Retrieval supplies context but does not persistently alter weights An active setting cannot independently prove the requested business outcome.
D:Similarity search alone is not a trained response style Design approval is not evidence that the alternative meets the actual requirement.
Question 244
A task is straightforward and the team wants the model to answer without supplying any examples. 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 only an embedding vector with no instruction
- A reviewed design record approves this competing approach: Assume an unrelated model evaluation result is the prompt
- The output follows the instruction without demonstration examples
- A system inventory lists this competing approach as active: Call the prompt few-shot despite containing no examples
Explanation
Correct Answer
C:The output follows the instruction without demonstration examples. This checks the business result rather than just a configured feature.
Incorrect Answers
A:A semantic vector is not a complete requested task A component invocation does not validate the necessary end-to-end outcome.
B:Evaluation metrics do not tell the model the current task Design approval is not evidence that the alternative meets the actual requirement.
D:Few-shot prompting includes demonstrations of the desired mapping An active setting cannot independently prove the requested business outcome.
Question 245
A travel assistant must answer in several regional languages with citations to local laws. Before approving a rollout, an architect suggests Pick by longest generated output without multilingual testing. Which mismatch should be documented for this precise design?
- Large completions are not proof of language coverage
- Cost savings cannot compensate for unusable translations
- Language quality and factual grounding are explicit selection criteria
- English performance does not prove regional language accuracy
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Pick by longest generated output without multilingual testing. Large completions are not proof of language coverage
Incorrect Answers
B:This limitation concerns Choose cheapest without examining linguistic accuracy; it does not diagnose the proposed Pick by longest generated output without multilingual testing approach.
C:Language quality and factual grounding are explicit selection criteria This explains the recommended design, but it is not the requested diagnosis of Pick by longest generated output without multilingual testing.
D:This limitation concerns Choose any text model solely by English benchmarks; it does not diagnose the proposed Pick by longest generated output without multilingual testing approach.
Question 246
A specialized science assistant requires substantial domain-language familiarity absent from generic pretraining. Which option BEST meets the business requirement?
- Assess continued pretraining or domain adaptation under cost and licensing constraints
- Use prompt caching as the only way to teach unfamiliar specialized language
- Treat a new user interface as domain adaptation
- Assume retrieval alone always yields the desired base domain language behavior
Explanation
Correct Answer
A:Further training on domain corpora can adapt foundational representations The specified scenario makes this the most appropriate approach.
Incorrect Answers
B:Caching reused prefixes does not train model weights This prevents the approach from meeting the stated requirement.
C:Presentation changes do not alter model language knowledge This prevents the approach from meeting the stated requirement.
D:RAG helps facts but may not substitute for deep domain adaptation This prevents the approach from meeting the stated requirement.
Question 247
An internal prompt includes confidential API keys that users might echo back. After selecting the recommended approach, which evidence MOST DIRECTLY demonstrates that the actual required outcome has been achieved?
- Prompt inspection finds no secrets and attempted tool access requires authorization
- A component test invokes this alternate approach successfully: Raise temperature so the model forgets the secret
- A system inventory lists this competing approach as active: Rely solely on hiding secret text in a longer prompt
- A reviewed design record approves this competing approach: Assume model refusal alone makes a leaked key invalid
Explanation
Correct Answer
A:Prompt inspection finds no secrets and attempted tool access requires authorization. This checks the business result rather than just a configured feature.
Incorrect Answers
B:Sampling randomness does not revoke exposed credentials A component invocation does not validate the necessary end-to-end outcome.
C:Obscurity is not credential protection An active setting cannot independently prove the requested business outcome.
D:Refusal behavior cannot replace credential rotation and least privilege Design approval is not evidence that the alternative meets the actual requirement.
Question 248
A summarization model must capture key ideas from reference summaries. A pilot uses this approach instead: Use only training loss to score final user-facing summaries. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Key configuration does not evaluate language output
- ROUGE compares overlap with reference text but misses some semantic quality
- Fast output may still omit or invent facts
- Training loss does not directly measure summary quality against a reference
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Use only training loss to score final user-facing summaries. Training loss does not directly measure summary quality against a reference
Incorrect Answers
A:This limitation concerns Use AWS KMS encryption status as a summary metric; it does not diagnose the proposed Use only training loss to score final user-facing summaries approach.
B:ROUGE compares overlap with reference text but misses some semantic quality This explains the recommended design, but it is not the requested diagnosis of Use only training loss to score final user-facing summaries.
C:This limitation concerns Treat low latency alone as summary faithfulness; it does not diagnose the proposed Use only training loss to score final user-facing summaries approach.
Question 249
A team repeatedly edits instructions used by three Bedrock applications and needs a common revision history. During a design review, someone proposes Change model temperature to create an audit history. Which technical objection is specifically relevant to this proposal?
- Audit logs record API activities rather than manage reusable prompts
- Sampling settings do not track who approved a prompt
- A managed prompt resource supports controlled prompt iteration and versions
- Ad hoc prompts lack shared version governance
Explanation
Correct Answer
B:This is the actual failure mode of the proposed approach: Change model temperature to create an audit history. Sampling settings do not track who approved a prompt
Incorrect Answers
A:This limitation concerns Write prompts into CloudTrail as the only prompt-template system; it does not diagnose the proposed Change model temperature to create an audit history approach.
C:A managed prompt resource supports controlled prompt iteration and versions This explains the recommended design, but it is not the requested diagnosis of Change model temperature to create an audit history.
D:This limitation concerns Store only untracked text snippets in personal chat histories; it does not diagnose the proposed Change model temperature to create an audit history approach.
Question 250
A task is straightforward and the team wants the model to answer without supplying any examples. Before approving a rollout, an architect suggests Assume an unrelated model evaluation result is the prompt. Which mismatch should be documented for this precise design?
- Few-shot prompting includes demonstrations of the desired mapping
- Evaluation metrics do not tell the model the current task
- A semantic vector is not a complete requested task
- A zero-shot prompt states the task without demonstrated examples
Explanation
Correct Answer
B:This is the actual failure mode of the proposed approach: Assume an unrelated model evaluation result is the prompt. Evaluation metrics do not tell the model the current task
Incorrect Answers
A:This limitation concerns Call the prompt few-shot despite containing no examples; it does not diagnose the proposed Assume an unrelated model evaluation result is the prompt approach.
C:This limitation concerns Use only an embedding vector with no instruction; it does not diagnose the proposed Assume an unrelated model evaluation result is the prompt approach.
D:A zero-shot prompt states the task without demonstrated examples This explains the recommended design, but it is not the requested diagnosis of Assume an unrelated model evaluation result is the prompt.
