AWS Certified AI Practitioner AIF-C01 Practice Test 12
Question 110
An insurance assistant must interpret an uploaded photo together with a text instruction. 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 audio transcription as image understanding
- A system inventory lists this competing approach as active: Use text-only inference without a visual description
- The assistant correctly cites visible details and follows text instructions
- A reviewed design record approves this competing approach: Use a filename-based regression score as the visual interpretation
Explanation
Correct Answer
C:The assistant correctly cites visible details and follows text instructions. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Speech recognition does not process picture content A component invocation does not validate the necessary end-to-end outcome.
B:Text-only input cannot directly inspect the photo An active setting cannot independently prove the requested business outcome.
D:File names are not evidence of damage in pixels Design approval is not evidence that the alternative meets the actual requirement.
Question 111
A retailer wants to show whether an assistant increases useful customer outcomes. During a design review, someone proposes Use model training accuracy as the sole business metric. Which technical objection is specifically relevant to this proposal?
- Training accuracy does not capture customer conversations
- Model size is not a financial outcome
- Business value requires outcome measures not only token throughput
- Speed alone does not measure customer resolution
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Use model training accuracy as the sole business metric. Training accuracy does not capture customer conversations
Incorrect Answers
B:This limitation concerns Treat parameter count as net revenue; it does not diagnose the proposed Use model training accuracy as the sole business metric approach.
C:Business value requires outcome measures not only token throughput This explains the recommended design, but it is not the requested diagnosis of Use model training accuracy as the sole business metric.
D:This limitation concerns Use tokens generated per second as the ROI metric; it does not diagnose the proposed Use model training accuracy as the sole business metric approach.
Question 112
An agent must access approved external APIs through a common tool interface. Before approving a rollout, an architect suggests Assume mentioning a tool in a prompt makes it callable. Which mismatch should be documented for this precise design?
- Similarity is not authorization
- MCP standardizes tool connectivity but does not grant permission automatically
- Breaking text into tokens does not authenticate API calls
- Prompt text cannot establish secure tool registration
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Assume mentioning a tool in a prompt makes it callable. Prompt text cannot establish secure tool registration
Incorrect Answers
A:This limitation concerns Use vector relevance as permission to transfer money; it does not diagnose the proposed Assume mentioning a tool in a prompt makes it callable approach.
B:MCP standardizes tool connectivity but does not grant permission automatically This explains the recommended design, but it is not the requested diagnosis of Assume mentioning a tool in a prompt makes it callable.
C:This limitation concerns Use tokenization as the API authorization mechanism; it does not diagnose the proposed Assume mentioning a tool in a prompt makes it callable approach.
Question 113
A product owner is choosing between text-only and image-capable models for photo-assisted replies. 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: Select by model popularity rather than interface capabilities
- The model accepts photos and correctly describes the pictured facts
- A system inventory lists this competing approach as active: Use text-only just because it is a foundation model
- A reviewed design record approves this competing approach: Assume language support automatically includes photos
Explanation
Correct Answer
B:The model accepts photos and correctly describes the pictured facts. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Popularity cannot guarantee image understanding A component invocation does not validate the necessary end-to-end outcome.
C:FM status does not imply image input An active setting cannot independently prove the requested business outcome.
D:Multilingual and multimodal support are independent Design approval is not evidence that the alternative meets the actual requirement.
Question 114
A design studio wants new product illustrations based on descriptions. A pilot uses this approach instead: Use text sentiment labels to paint the image. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Sentiment does not produce synthetic images
- OCR reads existing images rather than creating new ones
- Generative visual models create new imagery from prompts
- A numeric prediction is not the requested visual artifact
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Use text sentiment labels to paint the image. Sentiment does not produce synthetic images
Incorrect Answers
B:This limitation concerns Use OCR to invent the artwork; it does not diagnose the proposed Use text sentiment labels to paint the image approach.
C:Generative visual models create new imagery from prompts This explains the recommended design, but it is not the requested diagnosis of Use text sentiment labels to paint the image.
D:This limitation concerns Use numeric regression as the illustration generator; it does not diagnose the proposed Use text sentiment labels to paint the image approach.
Question 115
A policy summarizer generates creative off-topic responses. A pilot uses this approach instead: Maximize temperature to guarantee faithfulness. Which limitation MOST DIRECTLY explains why the requirement is not met?
- High randomness increases diversity rather than factual assurance
- Lack of evidence increases unsupported-answer risk
- More generated tokens do not ensure relevance
- Lower sampling randomness can reduce variation but does not replace grounding
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Maximize temperature to guarantee faithfulness. High randomness increases diversity rather than factual assurance
Incorrect Answers
B:This limitation concerns Remove all source context; it does not diagnose the proposed Maximize temperature to guarantee faithfulness approach.
C:This limitation concerns Increase output length without evaluating quality; it does not diagnose the proposed Maximize temperature to guarantee faithfulness approach.
D:Lower sampling randomness can reduce variation but does not replace grounding This explains the recommended design, but it is not the requested diagnosis of Maximize temperature to guarantee faithfulness.
Question 116
A customer expects identical exact wording for every creative-answer request. 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: Promise byte-identical text from every call
- A component test invokes this alternate approach successfully: Assume longer prompts alone guarantee equality
- Repeated tests assess factual consistency and allowed variation
- A reviewed design record approves this competing approach: Add an embedding database to force identical prose
Explanation
Correct Answer
C:Repeated tests assess factual consistency and allowed variation. This checks the business result rather than just a configured feature.
Incorrect Answers
A:Generative outputs are not universally deterministic An active setting cannot independently prove the requested business outcome.
B:Length is not a reproducibility guarantee A component invocation does not validate the necessary end-to-end outcome.
D:Retrieval does not force identical generation Design approval is not evidence that the alternative meets the actual requirement.
Question 117
A support chatbot drops critical policy sections when every long document is pasted into its prompt. Before approving a rollout, an architect suggests Use uppercase characters to increase context size. Which mismatch should be documented for this precise design?
- Context windows limit how much text a model can accept
- Casing does not expand the model's token limit
- Sampling temperature does not enlarge prompt capacity
- Duplication consumes capacity and can crowd out evidence
Explanation
Correct Answer
B:This is the actual failure mode of the proposed approach: Use uppercase characters to increase context size. Casing does not expand the model's token limit
Incorrect Answers
A:Context windows limit how much text a model can accept This explains the recommended design, but it is not the requested diagnosis of Use uppercase characters to increase context size.
C:This limitation concerns Raise temperature to expand the context window; it does not diagnose the proposed Use uppercase characters to increase context size approach.
D:This limitation concerns Repeat each policy passage several times; it does not diagnose the proposed Use uppercase characters to increase context size approach.
Question 118
An assistant gives convincing but invented policy clauses. A pilot uses this approach instead: Assume fluent wording guarantees factual accuracy. Which limitation MOST DIRECTLY explains why the requirement is not met?
- Higher randomness cannot create missing evidence
- Long responses may contain unsupported content
- Fluent LLM output can still be false
- Style is not factual evidence
Explanation
Correct Answer
D:This is the actual failure mode of the proposed approach: Assume fluent wording guarantees factual accuracy. Style is not factual evidence
Incorrect Answers
A:This limitation concerns Increase sampling randomness to ensure correct facts; it does not diagnose the proposed Assume fluent wording guarantees factual accuracy approach.
B:This limitation concerns Treat output length as a correctness certificate; it does not diagnose the proposed Assume fluent wording guarantees factual accuracy approach.
C:Fluent LLM output can still be false This explains the recommended design, but it is not the requested diagnosis of Assume fluent wording guarantees factual accuracy.
Question 119
A team wants managed access to several pretrained FMs through an AWS API. Before approving a rollout, an architect suggests Treat an S3 object as an automatic model runtime. Which mismatch should be documented for this precise design?
- Bedrock provides managed access to supported FMs
- CloudTrail logs account activity rather than generating text
- Storage does not by itself run FM inference
- Textract focuses on document extraction
Explanation
Correct Answer
C:This is the actual failure mode of the proposed approach: Treat an S3 object as an automatic model runtime. Storage does not by itself run FM inference
Incorrect Answers
A:Bedrock provides managed access to supported FMs This explains the recommended design, but it is not the requested diagnosis of Treat an S3 object as an automatic model runtime.
B:This limitation concerns Use AWS CloudTrail as the inference endpoint; it does not diagnose the proposed Treat an S3 object as an automatic model runtime approach.
D:This limitation concerns Use Amazon Textract as the FM catalog; it does not diagnose the proposed Treat an S3 object as an automatic model runtime approach.
Question 120
A knowledge bot needs to retrieve a particular paragraph from a long manual. Before approving a rollout, an architect suggests Increase temperature instead of changing chunk boundaries. Which mismatch should be documented for this precise design?
- Generation randomness does not fix coarse retrieval
- Local chunks improve retrieval granularity and provenance
- Missing provenance makes source verification difficult
- Whole-document representations can obscure local answers
Explanation
Correct Answer
A:This is the actual failure mode of the proposed approach: Increase temperature instead of changing chunk boundaries. Generation randomness does not fix coarse retrieval
Incorrect Answers
B:Local chunks improve retrieval granularity and provenance This explains the recommended design, but it is not the requested diagnosis of Increase temperature instead of changing chunk boundaries.
C:This limitation concerns Delete document identifiers during ingestion; it does not diagnose the proposed Increase temperature instead of changing chunk boundaries approach.
D:This limitation concerns Embed each manual as one undivided record; it does not diagnose the proposed Increase temperature instead of changing chunk boundaries approach.
