Amazon AWS Certified AI Practitioner AIF-C01 Practice Test 11
Question 100
The initial scenario was: An IT team compares consuming managed inference from an API with hosting its own model-serving runtime. The original control is already in place. Now the product later needs a fully custom model trained on proprietary labeled examples. Which ADDITIONAL action BEST addresses the new condition?
- Compare managed API operational simplicity with self-hosting control and maintenance
- A workload decision records cost, latency and operational ownership in a pilot
- Assume managed APIs and self-hosting carry identical operational duties
- Evaluate a managed training workflow in SageMaker AI
Explanation
Correct Answer
D:The product later needs a fully custom model trained on proprietary labeled examples. The appropriate added action is: Evaluate a managed training workflow in SageMaker AI.
Incorrect Answers
A:This was the original control and is already implemented. Managed inference reduces infrastructure ownership while self-hosting may require patching and scaling.
B:This describes evidence for the original approach, not a new action: A workload decision records cost, latency and operational ownership in a pilot.
C:This original shortcut remains unsuitable: The hosting choice changes infrastructure and maintenance responsibility.
Question 101
A support assistant's usage report bills for input and generated output even though the request contains no images. Which option BEST meets the stated AI/ML or AWS requirement?
- Treat each complete conversation as exactly one fixed token
- Shorten repetitive context while preserving needed instructions
- Divide documents into coherent retrievable chunks
- Explain input and output tokens as usage units for model processing
Explanation
Correct Answer
D:Token counts for prompts and responses commonly drive FM inference usage charges. In this situation, choose: Explain input and output tokens as usage units for model processing.
Incorrect Answers
A:A conversation can contain many tokens across its context and generated text. The proposal, Treat each complete conversation as exactly one fixed token, does not satisfy this need.
B:This is appropriate only after a distinct change: The team later sees overly long prompts increase the bill. It is not the best initial response to a support assistant's usage report bills for input and generated output even though the request contains no images.
C:This can be valid for a different task: A team indexes 600-page manuals so retrieval finds relevant paragraphs rather than sending every page into a model. The current request is different and needs the selected approach.
Question 102
A support assistant's usage report bills for input and generated output even though the request contains no images. A colleague suggests: Treat each complete conversation as exactly one fixed token. Which finding BEST explains why the suggestion should be rejected?
- Giant documents are difficult to search and may exceed model context limits
- Lexical matching alone can miss semantically equivalent queries
- The shortcut is safe because its terminology is familiar to the team
- A conversation can contain many tokens across its context and generated text
Explanation
Correct Answer
D:A conversation can contain many tokens across its context and generated text. This is the direct defect in the proposed action: Treat each complete conversation as exactly one fixed token.
Incorrect Answers
A:This finding is associated with a different problem: A team indexes 600-page manuals so retrieval finds relevant paragraphs rather than sending every page into a model.
B:This concern belongs to another situation: A product-support search returns relevant articles despite customers using different words than the documentation.
C:Familiar wording is not proof that an action is appropriate. The actual defect is: A conversation can contain many tokens across its context and generated text.
Question 103
A support assistant's usage report bills for input and generated output even though the request contains no images. After implementing Explain input and output tokens as usage units for model processing, which observation BEST verifies the intended result?
- The team approves Explain input and output tokens as usage units for model processing but does not run an operational test
- The team schedules a future test of whether monitoring shows input and output token totals on representative requests, but has no result
- Monitoring shows input and output token totals on representative requests
- A query retrieves the specific relevant manual sections instead of whole files
Explanation
Correct Answer
C:The observed result, Monitoring shows input and output token totals on representative requests, provides direct evidence. Token counts for prompts and responses commonly drive FM inference usage charges.
Incorrect Answers
A:The required observable condition is: Monitoring shows input and output token totals on representative requests.
B:Verify the actual outcome: Monitoring shows input and output token totals on representative requests.
D:This observation would be evidence for another case: A team indexes 600-page manuals so retrieval finds relevant paragraphs rather than sending every page into a model, not the current scenario.
Question 104
The initial scenario was: A support assistant's usage report bills for input and generated output even though the request contains no images. The original control is already in place. Now the team later sees overly long prompts increase the bill. Which ADDITIONAL action BEST addresses the new condition?
- Explain input and output tokens as usage units for model processing
- Monitoring shows input and output token totals on representative requests
- Treat each complete conversation as exactly one fixed token
- Shorten repetitive context while preserving needed instructions
Explanation
Correct Answer
D:The team later sees overly long prompts increase the bill. The appropriate added action is: Shorten repetitive context while preserving needed instructions.
Incorrect Answers
A:This was the original control and is already implemented. Token counts for prompts and responses commonly drive FM inference usage charges.
B:This describes evidence for the original approach, not a new action: Monitoring shows input and output token totals on representative requests.
C:This original shortcut remains unsuitable: A conversation can contain many tokens across its context and generated text.
Question 105
A team indexes 600-page manuals so retrieval finds relevant paragraphs rather than sending every page into a model. Which option BEST meets the stated AI/ML or AWS requirement?
- Divide documents into coherent retrievable chunks
- Preserve chunk-to-source metadata and return citations
- Store every long manual only as one indivisible context item
- Use semantic embeddings to represent meaning in vector space
Explanation
Correct Answer
A:Chunking creates smaller passages suitable for embedding and retrieval. In this situation, choose: Divide documents into coherent retrievable chunks.
Incorrect Answers
B:This is appropriate only after a distinct change: Users ask for source attribution with every generated answer. It is not the best initial response to a team indexes 600-page manuals so retrieval finds relevant paragraphs rather than sending every page into a model.
C:Giant documents are difficult to search and may exceed model context limits. The proposal, Store every long manual only as one indivisible context item, does not satisfy this need.
D:This can be valid for a different task: A product-support search returns relevant articles despite customers using different words than the documentation. The current request is different and needs the selected approach.
Question 106
A team indexes 600-page manuals so retrieval finds relevant paragraphs rather than sending every page into a model. A colleague suggests: Store every long manual only as one indivisible context item. Which finding BEST explains why the suggestion should be rejected?
- Lexical matching alone can miss semantically equivalent queries
- Creation order does not represent meaning similarity
- The shortcut is safe because its terminology is familiar to the team
- Giant documents are difficult to search and may exceed model context limits
Explanation
Correct Answer
D:Giant documents are difficult to search and may exceed model context limits. This is the direct defect in the proposed action: Store every long manual only as one indivisible context item.
Incorrect Answers
A:This finding is associated with a different problem: A product-support search returns relevant articles despite customers using different words than the documentation.
B:This concern belongs to another situation: A data engineer asks why a database needs vector similarity when implementing semantic search.
C:Familiar wording is not proof that an action is appropriate. The actual defect is: Giant documents are difficult to search and may exceed model context limits.
Question 107
A team indexes 600-page manuals so retrieval finds relevant paragraphs rather than sending every page into a model. After implementing Divide documents into coherent retrievable chunks, which observation BEST verifies the intended result?
- The team schedules a future test of whether a query retrieves the specific relevant manual sections instead of whole files, but has no result
- The team approves Divide documents into coherent retrievable chunks but does not run an operational test
- A query retrieves the specific relevant manual sections instead of whole files
- Queries with synonyms retrieve meaning-related articles in an evaluated sample
Explanation
Correct Answer
C:The observed result, A query retrieves the specific relevant manual sections instead of whole files, provides direct evidence. Chunking creates smaller passages suitable for embedding and retrieval.
Incorrect Answers
A:Verify the actual outcome: A query retrieves the specific relevant manual sections instead of whole files.
B:The required observable condition is: A query retrieves the specific relevant manual sections instead of whole files.
D:This observation would be evidence for another case: A product-support search returns relevant articles despite customers using different words than the documentation, not the current scenario.
Question 108
The initial scenario was: A team indexes 600-page manuals so retrieval finds relevant paragraphs rather than sending every page into a model. The original control is already in place. Now users ask for source attribution with every generated answer. Which ADDITIONAL action BEST addresses the new condition?
- Preserve chunk-to-source metadata and return citations
- Divide documents into coherent retrievable chunks
- A query retrieves the specific relevant manual sections instead of whole files
- Store every long manual only as one indivisible context item
Explanation
Correct Answer
A:Users ask for source attribution with every generated answer. The appropriate added action is: Preserve chunk-to-source metadata and return citations.
Incorrect Answers
B:This was the original control and is already implemented. Chunking creates smaller passages suitable for embedding and retrieval.
C:This describes evidence for the original approach, not a new action: A query retrieves the specific relevant manual sections instead of whole files.
D:This original shortcut remains unsuitable: Giant documents are difficult to search and may exceed model context limits.
Question 109
A product-support search returns relevant articles despite customers using different words than the documentation. Which option BEST meets the stated AI/ML or AWS requirement?
- Use semantic embeddings to represent meaning in vector space
- Compare query embeddings with stored document vectors using similarity search
- Restrict retrieval to exact raw-string character matches only
- Attach retrievable document identifiers and show citations
Explanation
Correct Answer
A:Similar meaning can be reflected in nearby embedding vectors even when keywords differ. In this situation, choose: Use semantic embeddings to represent meaning in vector space.
Incorrect Answers
B:This can be valid for a different task: A data engineer asks why a database needs vector similarity when implementing semantic search. The current request is different and needs the selected approach.
C:Lexical matching alone can miss semantically equivalent queries. The proposal, Restrict retrieval to exact raw-string character matches only, does not satisfy this need.
D:This is appropriate only after a distinct change: Support agents demand authoritative citations instead of ungrounded summaries. It is not the best initial response to a product-support search returns relevant articles despite customers using different words than the documentation.
Question 110
A product-support search returns relevant articles despite customers using different words than the documentation. A colleague suggests: Restrict retrieval to exact raw-string character matches only. Which finding BEST explains why the suggestion should be rejected?
- Creation order does not represent meaning similarity
- Lexical matching alone can miss semantically equivalent queries
- Text classification cannot create the requested images
- The shortcut is safe because its terminology is familiar to the team
Explanation
Correct Answer
B:Lexical matching alone can miss semantically equivalent queries. This is the direct defect in the proposed action: Restrict retrieval to exact raw-string character matches only.
Incorrect Answers
A:This finding is associated with a different problem: A data engineer asks why a database needs vector similarity when implementing semantic search.
C:This concern belongs to another situation: A creative agency needs an AI model that produces new images from a written description.
D:Familiar wording is not proof that an action is appropriate. The actual defect is: Lexical matching alone can miss semantically equivalent queries.