Amazon AWS Certified AI Practitioner AIF-C01 Practice Test 31
Question 300
The initial scenario was: A business task needs a one-paragraph summary and the prompt includes ten pages of unrelated chat logs. The original control is already in place. Now a follow-up asks for citations to exact policy passages. Which ADDITIONAL action BEST addresses the new condition?
- Retrieve and attach source references to the prompt context
- Summaries meet length constraints and accurately reflect the source text
- Include every unrelated archive message even after context overflow
- Keep task-relevant context and specify the summary length explicitly
Explanation
Correct Answer
A:A follow-up asks for citations to exact policy passages. The appropriate added action is: Retrieve and attach source references to the prompt context.
Incorrect Answers
B:This describes evidence for the original approach, not a new action: Summaries meet length constraints and accurately reflect the source text.
C:This original shortcut remains unsuitable: Excess irrelevant text is not a substitute for targeted task evidence.
D:This was the original control and is already implemented. Irrelevant context consumes tokens and can divert attention from the requested summary.
Question 301
A team wants the model to distinguish authorized application instructions from user-provided text that imitates an administrator command. Which option BEST meets the stated AI/ML or AWS requirement?
- Identify the approach as zero-shot prompting
- Separate trusted instruction layers and restrict user text to its intended data role
- Add identity-based authorization and explicit confirmation for approved actions
- Treat the words 'SYSTEM OVERRIDE' in a user message as actual system authority
Explanation
Correct Answer
B:Text that claims higher authority does not gain permission by being in a prompt. In this situation, choose: Separate trusted instruction layers and restrict user text to its intended data role.
Incorrect Answers
A:This can be valid for a different task: A new prompt asks a general-purpose model to categorize a support message without providing any examples. The current request is different and needs the selected approach.
C:This is appropriate only after a distinct change: The app later permits limited actions requested by authenticated staff. It is not the best initial response to a team wants the model to distinguish authorized application instructions from user-provided text that imitates an administrator command.
D:User-authored role claims cannot override trusted policy or tool scopes. The proposal, Treat the words 'SYSTEM OVERRIDE' in a user message as actual system authority, does not satisfy this need.
Question 302
A team wants the model to distinguish authorized application instructions from user-provided text that imitates an administrator command. A colleague suggests: Treat the words 'SYSTEM OVERRIDE' in a user message as actual system authority. Which finding BEST explains why the suggestion should be rejected?
- Few-shot prompting includes examples in the prompt
- The shortcut is safe because its terminology is familiar to the team
- In-context examples condition inference without training parameter updates
- User-authored role claims cannot override trusted policy or tool scopes
Explanation
Correct Answer
D:User-authored role claims cannot override trusted policy or tool scopes. This is the direct defect in the proposed action: Treat the words 'SYSTEM OVERRIDE' in a user message as actual system authority.
Incorrect Answers
A:This finding is associated with a different problem: A new prompt asks a general-purpose model to categorize a support message without providing any examples.
B:Familiar wording is not proof that an action is appropriate. The actual defect is: User-authored role claims cannot override trusted policy or tool scopes.
C:This concern belongs to another situation: A sentiment task works poorly with bare instructions, so the prompt includes three labeled example reviews.
Question 303
A team wants the model to distinguish authorized application instructions from user-provided text that imitates an administrator command. After implementing Separate trusted instruction layers and restrict user text to its intended data role, which observation BEST verifies the intended result?
- Adversarial prompts fail to execute prohibited tool actions in tests
- The team approves Separate trusted instruction layers and restrict user text to its intended data role but does not run an operational test
- A test response follows the requested categories without supplied samples
- The team schedules a future test of whether adversarial prompts fail to execute prohibited tool actions in tests, but has no result
Explanation
Correct Answer
A:The observed result, Adversarial prompts fail to execute prohibited tool actions in tests, provides direct evidence. Text that claims higher authority does not gain permission by being in a prompt.
Incorrect Answers
B:The required observable condition is: Adversarial prompts fail to execute prohibited tool actions in tests.
C:This observation would be evidence for another case: A new prompt asks a general-purpose model to categorize a support message without providing any examples, not the current scenario.
D:An unperformed future check is not evidence. Verify the actual outcome: Adversarial prompts fail to execute prohibited tool actions in tests.
Question 304
The initial scenario was: A team wants the model to distinguish authorized application instructions from user-provided text that imitates an administrator command. The original control is already in place. Now the app later permits limited actions requested by authenticated staff. Which ADDITIONAL action BEST addresses the new condition?
- Treat the words 'SYSTEM OVERRIDE' in a user message as actual system authority
- Add identity-based authorization and explicit confirmation for approved actions
- Separate trusted instruction layers and restrict user text to its intended data role
- Adversarial prompts fail to execute prohibited tool actions in tests
Explanation
Correct Answer
B:The app later permits limited actions requested by authenticated staff. The appropriate added action is: Add identity-based authorization and explicit confirmation for approved actions.
Incorrect Answers
A:This original shortcut remains unsuitable: User-authored role claims cannot override trusted policy or tool scopes.
C:This was the original control and is already implemented. Text that claims higher authority does not gain permission by being in a prompt.
D:This describes evidence for the original approach, not a new action: Adversarial prompts fail to execute prohibited tool actions in tests.
Question 305
A company asks whether a general language model can be pretrained on millions of diverse documents before specializing it. Which option BEST meets the stated AI/ML or AWS requirement?
- Evaluate instruction fine-tuning with representative approved examples
- Describe large-scale pretraining as learning broad patterns from extensive data
- Call pretraining a ten-example prompt with no parameter updates
- Evaluate instruction fine-tuning or prompts for the behavior change
Explanation
Correct Answer
B:Pretraining builds general model representations before downstream adaptation. In this situation, choose: Describe large-scale pretraining as learning broad patterns from extensive data.
Incorrect Answers
A:This can be valid for a different task: An enterprise has curated customer-service dialogues and wants a pretrained model to follow its response protocol. The current request is different and needs the selected approach.
C:In-context examples are not large-scale weight-training pretraining. The proposal, Call pretraining a ten-example prompt with no parameter updates, does not satisfy this need.
D:This is appropriate only after a distinct change: The product later needs to follow a specialized output style. It is not the best initial response to a company asks whether a general language model can be pretrained on millions of diverse documents before specializing it.
Question 306
A company asks whether a general language model can be pretrained on millions of diverse documents before specializing it. A colleague suggests: Call pretraining a ten-example prompt with no parameter updates. Which finding BEST explains why the suggestion should be rejected?
- The shortcut is safe because its terminology is familiar to the team
- Uncurated unrelated data is unlikely to teach the required instructions
- Unreviewed sensitive data creates confidentiality and compliance risks
- In-context examples are not large-scale weight-training pretraining
Explanation
Correct Answer
D:In-context examples are not large-scale weight-training pretraining. This is the direct defect in the proposed action: Call pretraining a ten-example prompt with no parameter updates.
Incorrect Answers
A:Familiar wording is not proof that an action is appropriate. The actual defect is: In-context examples are not large-scale weight-training pretraining.
B:This finding is associated with a different problem: An enterprise has curated customer-service dialogues and wants a pretrained model to follow its response protocol.
C:This concern belongs to another situation: A healthcare organization evaluates fine-tuning on internal case records containing patient information.
Question 307
A company asks whether a general language model can be pretrained on millions of diverse documents before specializing it. After implementing Describe large-scale pretraining as learning broad patterns from extensive data, which observation BEST verifies the intended result?
- Held-out dialogue examples show improved compliance with service response rules
- The team approves Describe large-scale pretraining as learning broad patterns from extensive data but does not run an operational test
- The team schedules a future test of whether documentation distinguishes the pretrained base model from later adapted versions, but has no result
- Documentation distinguishes the pretrained base model from later adapted versions
Explanation
Correct Answer
D:The observed result, Documentation distinguishes the pretrained base model from later adapted versions, provides direct evidence. Pretraining builds general model representations before downstream adaptation.
Incorrect Answers
A:This observation would be evidence for another case: An enterprise has curated customer-service dialogues and wants a pretrained model to follow its response protocol, not the current scenario.
B:The required observable condition is: Documentation distinguishes the pretrained base model from later adapted versions.
C:Verify the actual outcome: Documentation distinguishes the pretrained base model from later adapted versions.
Question 308
The initial scenario was: A company asks whether a general language model can be pretrained on millions of diverse documents before specializing it. The original control is already in place. Now the product later needs to follow a specialized output style. Which ADDITIONAL action BEST addresses the new condition?
- Evaluate instruction fine-tuning or prompts for the behavior change
- Call pretraining a ten-example prompt with no parameter updates
- Documentation distinguishes the pretrained base model from later adapted versions
- Describe large-scale pretraining as learning broad patterns from extensive data
Explanation
Correct Answer
A:The product later needs to follow a specialized output style. The appropriate added action is: Evaluate instruction fine-tuning or prompts for the behavior change.
Incorrect Answers
B:This original shortcut remains unsuitable: In-context examples are not large-scale weight-training pretraining.
C:This describes evidence for the original approach, not a new action: Documentation distinguishes the pretrained base model from later adapted versions.
D:This was the original control and is already implemented. Pretraining builds general model representations before downstream adaptation.
Question 309
An enterprise has curated customer-service dialogues and wants a pretrained model to follow its response protocol. Which option BEST meets the stated AI/ML or AWS requirement?
- Add retrieval over up-to-date knowledge instead of retraining for every update
- Curate, de-identify and govern the training set before customization
- Evaluate instruction fine-tuning with representative approved examples
- Train only on unrelated unlabeled log noise and expect exact protocol compliance
Explanation
Correct Answer
C:Instruction tuning adapts model behavior using task-aligned training data. In this situation, choose: Evaluate instruction fine-tuning with representative approved examples.
Incorrect Answers
A:This is appropriate only after a distinct change: Knowledge facts change daily in the support database. It is not the best initial response to an enterprise has curated customer-service dialogues and wants a pretrained model to follow its response protocol.
B:This can be valid for a different task: A healthcare organization evaluates fine-tuning on internal case records containing patient information. The current request is different and needs the selected approach.
D:Uncurated unrelated data is unlikely to teach the required instructions. The proposal, Train only on unrelated unlabeled log noise and expect exact protocol compliance, does not satisfy this need.
Question 310
An enterprise has curated customer-service dialogues and wants a pretrained model to follow its response protocol. A colleague suggests: Train only on unrelated unlabeled log noise and expect exact protocol compliance. Which finding BEST explains why the suggestion should be rejected?
- Inference prompts do not by themselves retrain model parameters
- Unreviewed sensitive data creates confidentiality and compliance risks
- Uncurated unrelated data is unlikely to teach the required instructions
- The shortcut is safe because its terminology is familiar to the team
Explanation
Correct Answer
C:Uncurated unrelated data is unlikely to teach the required instructions. This is the direct defect in the proposed action: Train only on unrelated unlabeled log noise and expect exact protocol compliance.
Incorrect Answers
A:This concern belongs to another situation: An AI team wants a foundation model to learn specialized vocabulary from a large corpus beyond simple response examples.
B:This finding is associated with a different problem: A healthcare organization evaluates fine-tuning on internal case records containing patient information.
D:Familiar wording is not proof that an action is appropriate. The actual defect is: Uncurated unrelated data is unlikely to teach the required instructions.