AWS Certified AI Practitioner AIF-C01 Practice Test 26
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.
Question 251
A large FM is too costly to serve, but a smaller model could learn some of its task behavior. Before approving a rollout, an architect suggests Replace the small student model with a larger untested teacher in production. Which mismatch should be documented for this precise design?
- Distillation can transfer useful behavior to a more efficient model
- More context does not reduce model parameters
- Retrieval indexing is not teacher-student model training
- A larger teacher alone does not solve serving-cost constraints
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Replace the small student model with a larger untested teacher in production. A larger teacher alone does not solve serving-cost constraints
Incorrect Answers
A:Distillation can transfer useful behavior to a more efficient model This explains the recommended design, but it is not the requested diagnosis of Replace the small student model with a larger untested teacher in production.
B:This limitation concerns Expand the context window as the entire model compression method; it does not diagnose the proposed Replace the small student model with a larger untested teacher in production approach.
C:This limitation concerns Assume distillation means indexing source PDFs in a vector DB; it does not diagnose the proposed Replace the small student model with a larger untested teacher in production approach.
Question 252
A retailer needs a recurring branded tone across thousands of responses based on a stable style guide. A pilot uses this approach instead: Assume RAG always changes the model's learned style permanently. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Weight adaptation can improve consistent behavior for a stable specialized style
- Retrieval supplies context but does not persistently alter weights
- Quota changes do not teach the desired tone
- Similarity search alone is not a trained response style
Explanation
Correct Answer
B:This is the actual failure mode of the proposed approach: Assume RAG always changes the model's learned style permanently. Retrieval supplies context but does not persistently alter weights
Incorrect Answers
A:Weight adaptation can improve consistent behavior for a stable specialized style This explains the recommended design, but it is not the requested diagnosis of Assume RAG always changes the model's learned style permanently.
C:This limitation concerns Use a token quota increase as the style training method; it does not diagnose the proposed Assume RAG always changes the model's learned style permanently approach.
D:This limitation concerns Rely only on an embedding vector index to enforce every phrasing choice; it does not diagnose the proposed Assume RAG always changes the model's learned style permanently approach.
Question 253
A travel assistant must answer in several regional languages with citations to local laws. 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: Pick by longest generated output without multilingual testing
- A component test invokes this alternate approach successfully: Choose cheapest without examining linguistic accuracy
- Native-language reviewers verify factual sourced answers across the required regions
- A system inventory lists this competing approach as active: Choose any text model solely by English benchmarks
Explanation
Correct Answer
C:Native-language reviewers verify factual sourced answers across the required regions. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Large completions are not proof of language coverage Design approval is not evidence that the alternative meets the actual requirement.
B:Cost savings cannot compensate for unusable translations A component invocation does not validate the necessary end-to-end outcome.
D:English performance does not prove regional language accuracy An active setting cannot independently prove the requested business outcome.
Question 254
A multi-step support agent appears fluent but regularly fails to complete the requested customer workflow. A pilot uses this approach instead: Score only conversational word count. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Longer responses can mask unfinished work
- Agent value depends on actual workflow outcomes, not only attractive prose
- Tool availability is not successful authorized execution
- Retrieval quality alone is not end-to-end task success
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Score only conversational word count. Longer responses can mask unfinished work
Incorrect Answers
B:Agent value depends on actual workflow outcomes, not only attractive prose This explains the recommended design, but it is not the requested diagnosis of Score only conversational word count.
C:This limitation concerns Treat configured tools as proof every action executed; it does not diagnose the proposed Score only conversational word count approach.
D:This limitation concerns Measure retrieval relevance without tracking whether actions finished; it does not diagnose the proposed Score only conversational word count approach.
Question 255
An internal prompt includes confidential API keys that users might echo back. Which option BEST meets the business requirement?
- Assume model refusal alone makes a leaked key invalid
- Rely solely on hiding secret text in a longer prompt
- Remove secrets from prompts and enforce access boundaries
- Raise temperature so the model forgets the secret
Explanation
Correct Answer
C:Prompts may be logged or exposed and should not contain embedded credentials The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Refusal behavior cannot replace credential rotation and least privilege This prevents the approach from meeting the stated requirement.
B:Obscurity is not credential protection This prevents the approach from meeting the stated requirement.
D:Sampling randomness does not revoke exposed credentials This prevents the approach from meeting the stated requirement.
Question 256
A model fine-tuned on a narrow support sample produces poor answers for a neglected customer segment. 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 model size as a fairness measure
- A reviewed design record approves this competing approach: Remove the holdout set to make training accuracy look better
- A system inventory lists this competing approach as active: Repeat the same narrow examples more often
- Held-out tests by segment show acceptable accuracy and no large unexplained gaps
Explanation
Correct Answer
D:Held-out tests by segment show acceptable accuracy and no large unexplained gaps. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Parameter count does not measure group performance A component invocation does not validate the necessary end-to-end outcome.
B:Removing evaluation hides generalization problems Design approval is not evidence that the alternative meets the actual requirement.
C:More duplicates do not add missing segment coverage An active setting cannot independently prove the requested business outcome.
Question 257
A large FM is too costly to serve, but a smaller model could learn some of its task behavior. 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: Replace the small student model with a larger untested teacher in production
- A system inventory lists this competing approach as active: Assume distillation means indexing source PDFs in a vector DB
- A component test invokes this alternate approach successfully: Expand the context window as the entire model compression method
- The distilled student reaches target quality at lower measured inference cost
Explanation
Correct Answer
D:The distilled student reaches target quality at lower measured inference cost. This checks the business result rather than just a configured feature.
Incorrect Answers
A:A larger teacher alone does not solve serving-cost constraints Design approval is not evidence that the alternative meets the actual requirement.
B:Retrieval indexing is not teacher-student model training An active setting cannot independently prove the requested business outcome.
C:More context does not reduce model parameters A component invocation does not validate the necessary end-to-end outcome.
Question 258
A model fine-tuned on a narrow support sample produces poor answers for a neglected customer segment. Which option BEST meets the business requirement?
- Remove the holdout set to make training accuracy look better
- Use only model size as a fairness measure
- Repeat the same narrow examples more often
- Expand representative training and evaluation examples across relevant populations
Explanation
Correct Answer
D:Coverage of real user scenarios reduces harmful specialization gaps The specified scenario makes this the most appropriate approach.
Incorrect Answers
A:Removing evaluation hides generalization problems This prevents the approach from meeting the stated requirement.
B:Parameter count does not measure group performance This prevents the approach from meeting the stated requirement.
C:More duplicates do not add missing segment coverage This prevents the approach from meeting the stated requirement.
Question 259
A team wants to train a smaller assistant using responses from a larger teacher model. A pilot uses this approach instead: Use a vector index as the student weights. Which limitation MOST DIRECTLY explains why the requirement is not met?
- A student can learn task behavior from stronger-model outputs
- Retrieval vectors do not replace student training
- Serving the teacher alone does not create a smaller trained model
- Prompt compression does not transfer learned parameters
Explanation
Correct Answer
B:This is the actual failure mode of the proposed approach: Use a vector index as the student weights. Retrieval vectors do not replace student training
Incorrect Answers
A:A student can learn task behavior from stronger-model outputs This explains the recommended design, but it is not the requested diagnosis of Use a vector index as the student weights.
C:This limitation concerns Give the same large teacher every request with no student; it does not diagnose the proposed Use a vector index as the student weights approach.
D:This limitation concerns Only compress prompt text and claim the model was distilled; it does not diagnose the proposed Use a vector index as the student weights approach.
Question 260
A retailer needs a recurring branded tone across thousands of responses based on a stable style guide. Before approving a rollout, an architect suggests Rely only on an embedding vector index to enforce every phrasing choice. Which mismatch should be documented for this precise design?
- Weight adaptation can improve consistent behavior for a stable specialized style
- Retrieval supplies context but does rather than persistently alter weights
- Quota changes do not teach the desired tone
- Similarity search alone is not a trained response style
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Rely only on an embedding vector index to enforce every phrasing choice. Similarity search alone is not a trained response style
Incorrect Answers
A:Weight adaptation can improve consistent behavior for a stable specialized style This explains the recommended design, but it is not the requested diagnosis of Rely only on an embedding vector index to enforce every phrasing choice.
B:This limitation concerns Assume RAG always changes the model's learned style permanently; it does not diagnose the proposed Rely only on an embedding vector index to enforce every phrasing choice approach.
C:This limitation concerns Use a token quota increase as the style training method; it does not diagnose the proposed Rely only on an embedding vector index to enforce every phrasing choice approach.
