McAfee-Secured Website

Exam Code: AI-300

Exam Name: Operationalizing Machine Learning and Generative AI Solutions

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Machine Learning Operations Engineer Associate

100% Microsoft AI-300 Guaranteed Success With Testking Training Materials

Get Real AI-300 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

213 Questions & Answers with Testing Engine

"Operationalizing Machine Learning and Generative AI Solutions Exam", also known as AI-300 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-300 Exam Engine. Your AI-300 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft AI-300 Practice Test 53

Question 520

A PTU plan assumes one capacity allocation can serve unlimited model types with no compatibility checks. What should the architect validate?

  1. Measure workload input and generated output tokens with representative request concurrency.
  2. Allocate at least the supported minimum PTU count for the chosen model and deployment type.
  3. Correlate request rate, token volume, utilization, and throttling metrics before capacity changes.
  4. Confirm eligible model, version, and deployment-type support for the planned allocation.

Explanation

Correct Answer

D:

Provisioned capacity is not universally interchangeable across every model and region without supported compatibility.

Incorrect Answers

A:

Measure workload input and generated output tokens with representative request concurrency. That operation would be relevant if: An engineer calculates required capacity from input token counts alone, ignoring long outputs. Its corresponding reason or verification is: Token mix and completion lengths affect effective processing throughput, not merely prompt-token volume. However the problem in this question is: A PTU plan assumes one capacity allocation can serve unlimited model types with no compatibility checks. The requirement instead calls for Confirm eligible model, version, and deployment-type support for the planned allocation; the question-specific check is: Provisioned capacity is not universally interchangeable across every model and region without supported compatibility.

B:

Allocate at least the supported minimum PTU count for the chosen model and deployment type. A separate source case supports that approach when: A deployment request supplies fewer PTUs than the model's documented minimum and provisioning rejects it. The supporting rationale or observable result is: Models and provisioned deployment classes can have minimum allocation requirements independent of current user traffic. This question describes a different issue: A PTU plan assumes one capacity allocation can serve unlimited model types with no compatibility checks. Address it through Confirm eligible model, version, and deployment-type support for the planned allocation, with the required justification or evidence: Provisioned capacity is not universally interchangeable across every model and region without supported compatibility.

C:

Correlate request rate, token volume, utilization, and throttling metrics before capacity changes. This is useful for the distinct issue: A deployed provisioned model shows high PTU utilization while some client calls receive throttling responses. The relevant technical rationale or evidence for that issue is: Together these measures indicate whether load distribution, token sizes, or insufficient capacity causes serving pressure. Here the reported issue is: A PTU plan assumes one capacity allocation can serve unlimited model types with no compatibility checks. The relevant response is: Confirm eligible model, version, and deployment-type support for the planned allocation. The first result cannot verify this different correction: Provisioned capacity is not universally interchangeable across every model and region without supported compatibility.

Question 521

A workload needs 24-hour availability but a team plans to delete and recreate the provisioned deployment each night to save money. What operational risk should be assessed?

  1. Call the exact provisioned deployment identifier associated with the allocated PTUs.
  2. Allocate at least the supported minimum PTU count for the chosen model and deployment type.
  3. Consider deployment startup, quota availability, and continuity before daily deprovisioning.
  4. Right-size the PTU allocation based on observed sustained workload demand.

Explanation

Correct Answer

C:

Repeated recreation can introduce downtime and uncertain capacity even if it reduces some billed hours.

Incorrect Answers

A:

The alternative, Call the exact provisioned deployment identifier associated with the allocated PTUs, belongs to a different technical condition: An operator creates a provisioned deployment and sends inference to a different deployment name, receiving an unexpected throughput limit. There the source-specific mechanism or proof is: Capacity allocations apply to specific deployments; traffic sent elsewhere cannot consume the intended reservation. The condition asked about here is: A workload needs 24-hour availability but a team plans to delete and recreate the provisioned deployment each night to save money. A suitable correction is Consider deployment startup, quota availability, and continuity before daily deprovisioning; the targeted justification or result is: Repeated recreation can introduce downtime and uncertain capacity even if it reduces some billed hours.

B:

Allocate at least the supported minimum PTU count for the chosen model and deployment type. That operation would be relevant if: A deployment request supplies fewer PTUs than the model's documented minimum and provisioning rejects it. Its corresponding reason or verification is: Models and provisioned deployment classes can have minimum allocation requirements independent of current user traffic. However the problem in this question is: A workload needs 24-hour availability but a team plans to delete and recreate the provisioned deployment each night to save money. The requirement instead calls for Consider deployment startup, quota availability, and continuity before daily deprovisioning; the question-specific check is: Repeated recreation can introduce downtime and uncertain capacity even if it reduces some billed hours.

D:

Right-size the PTU allocation based on observed sustained workload demand. A separate source case supports that approach when: A provisioned deployment consistently shows very low utilization while consuming its full hourly allocation. The supporting rationale or observable result is: Low utilization can signal overprovisioning that increases cost without providing meaningful user benefit. This question describes a different issue: A workload needs 24-hour availability but a team plans to delete and recreate the provisioned deployment each night to save money. Address it through Consider deployment startup, quota availability, and continuity before daily deprovisioning, with the required justification or evidence: Repeated recreation can introduce downtime and uncertain capacity even if it reduces some billed hours.

Question 522

Traffic forecasting for a provisioned model uses daily average requests and ignores synchronized Monday-morning bursts. What should be included?

  1. Allocate at least the supported minimum PTU count for the chosen model and deployment type.
  2. Size PTUs using realistic peak concurrency and token rate, not only daily averages.
  3. Correlate request rate, token volume, utilization, and throttling metrics before capacity changes.
  4. Remeasure token throughput and latency on the exact new model version.

Explanation

Correct Answer

B:

Capacity planning must handle short high-load windows that average traffic statistics can conceal.

Incorrect Answers

A:

Allocate at least the supported minimum PTU count for the chosen model and deployment type. It addresses another case: A deployment request supplies fewer PTUs than the model's documented minimum and provisioning rejects it. For that case, the distinguishing justification or acceptance check is: Models and provisioned deployment classes can have minimum allocation requirements independent of current user traffic. The present case instead states: Traffic forecasting for a provisioned model uses daily average requests and ignores synchronized Monday-morning bursts. It calls for Size PTUs using realistic peak concurrency and token rate, not only daily averages. Showing success for the other case would not establish: Capacity planning must handle short high-load windows that average traffic statistics can conceal.

C:

The alternative, Correlate request rate, token volume, utilization, and throttling metrics before capacity changes, belongs to a different technical condition: A deployed provisioned model shows high PTU utilization while some client calls receive throttling responses. There the source-specific mechanism or proof is: Together these measures indicate whether load distribution, token sizes, or insufficient capacity causes serving pressure. The condition asked about here is: Traffic forecasting for a provisioned model uses daily average requests and ignores synchronized Monday-morning bursts. A suitable correction is Size PTUs using realistic peak concurrency and token rate, not only daily averages; the targeted justification or result is: Capacity planning must handle short high-load windows that average traffic statistics can conceal.

D:

Remeasure token throughput and latency on the exact new model version. That operation would be relevant if: A model provider changes the model version and its token processing profile. The old PTU estimate is reused without testing. Its corresponding reason or verification is: Provisioned capacity behavior depends on model characteristics; prior measurements may not transfer across versions. However the problem in this question is: Traffic forecasting for a provisioned model uses daily average requests and ignores synchronized Monday-morning bursts. The requirement instead calls for Size PTUs using realistic peak concurrency and token rate, not only daily averages; the question-specific check is: Capacity planning must handle short high-load windows that average traffic statistics can conceal.

Question 523

A company wants a recovery model endpoint if a primary provisioned deployment reaches capacity, but its alternative has different data-handling terms. What should failover design verify?

  1. Validate approved routing, model equivalence, and geographic policies for overflow handling.
  2. Call the exact provisioned deployment identifier associated with the allocated PTUs.
  3. Measure workload input and generated output tokens with representative request concurrency.
  4. Size PTUs using realistic peak concurrency and token rate, not only daily averages.

Explanation

Correct Answer

A:

Failover must preserve required security and behavior rather than sending requests to any available model.

Incorrect Answers

B:

Call the exact provisioned deployment identifier associated with the allocated PTUs. This is useful for the distinct issue: An operator creates a provisioned deployment and sends inference to a different deployment name, receiving an unexpected throughput limit. The relevant technical rationale or evidence for that issue is: Capacity allocations apply to specific deployments; traffic sent elsewhere cannot consume the intended reservation. Here the reported issue is: A company wants a recovery model endpoint if a primary provisioned deployment reaches capacity, but its alternative has different data-handling terms. The relevant response is: Validate approved routing, model equivalence, and geographic policies for overflow handling. The first result cannot verify this different correction: Failover must preserve required security and behavior rather than sending requests to any available model.

C:

Measure workload input and generated output tokens with representative request concurrency. It addresses another case: An engineer calculates required capacity from input token counts alone, ignoring long outputs. For that case, the distinguishing justification or acceptance check is: Token mix and completion lengths affect effective processing throughput, not merely prompt-token volume. The present case instead states: A company wants a recovery model endpoint if a primary provisioned deployment reaches capacity, but its alternative has different data-handling terms. It calls for Validate approved routing, model equivalence, and geographic policies for overflow handling. Showing success for the other case would not establish: Failover must preserve required security and behavior rather than sending requests to any available model.

D:

The alternative, Size PTUs using realistic peak concurrency and token rate, not only daily averages, belongs to a different technical condition: Traffic forecasting for a provisioned model uses daily average requests and ignores synchronized Monday-morning bursts. There the source-specific mechanism or proof is: Capacity planning must handle short high-load windows that average traffic statistics can conceal. The condition asked about here is: A company wants a recovery model endpoint if a primary provisioned deployment reaches capacity, but its alternative has different data-handling terms. A suitable correction is Validate approved routing, model equivalence, and geographic policies for overflow handling; the targeted justification or result is: Failover must preserve required security and behavior rather than sending requests to any available model.

Question 524

A deployed provisioned model shows high PTU utilization while some client calls receive throttling responses. What should the team examine?

  1. Confirm eligible model, version, and deployment-type support for the planned allocation.
  2. Measure workload input and generated output tokens with representative request concurrency.
  3. Request approved regional PTU quota or select a supported alternative deployment location.
  4. Correlate request rate, token volume, utilization, and throttling metrics before capacity changes.

Explanation

Correct Answer

D:

Together these measures indicate whether load distribution, token sizes, or insufficient capacity causes serving pressure.

Incorrect Answers

A:

Confirm eligible model, version, and deployment-type support for the planned allocation. A separate source case supports that approach when: A PTU plan assumes one capacity allocation can serve unlimited model types with no compatibility checks. The supporting rationale or observable result is: Provisioned capacity is not universally interchangeable across every model and region without supported compatibility. This question describes a different issue: A deployed provisioned model shows high PTU utilization while some client calls receive throttling responses. Address it through Correlate request rate, token volume, utilization, and throttling metrics before capacity changes, with the required justification or evidence: Together these measures indicate whether load distribution, token sizes, or insufficient capacity causes serving pressure.

B:

Measure workload input and generated output tokens with representative request concurrency. This is useful for the distinct issue: An engineer calculates required capacity from input token counts alone, ignoring long outputs. The relevant technical rationale or evidence for that issue is: Token mix and completion lengths affect effective processing throughput, not merely prompt-token volume. Here the reported issue is: A deployed provisioned model shows high PTU utilization while some client calls receive throttling responses. The relevant response is: Correlate request rate, token volume, utilization, and throttling metrics before capacity changes. The first result cannot verify this different correction: Together these measures indicate whether load distribution, token sizes, or insufficient capacity causes serving pressure.

C:

Request approved regional PTU quota or select a supported alternative deployment location. It addresses another case: A team requests a provisioned model in a region with no PTU quota available. For that case, the distinguishing justification or acceptance check is: Provisioned deployment capacity is constrained by actual quota and availability for the chosen region and model. The present case instead states: A deployed provisioned model shows high PTU utilization while some client calls receive throttling responses. It calls for Correlate request rate, token volume, utilization, and throttling metrics before capacity changes. Showing success for the other case would not establish: Together these measures indicate whether load distribution, token sizes, or insufficient capacity causes serving pressure.

Question 525

A prompt includes a customer message containing the text ignore previous instructions, and the model treats it as policy. How should the prompt structure change?

  1. Parameterize the prompt's document type and task-specific context fields.
  2. Require confirmation of transactional prerequisites before making the tool call.
  3. Define the output schema and constrain responses to the required fields.
  4. Clearly separate untrusted customer content from trusted system-level instructions.

Explanation

Correct Answer

D:

Instruction hierarchy and data delimiters reduce the chance that a retrieved or user-supplied string overrides the application's policy.

Incorrect Answers

A:

Parameterize the prompt's document type and task-specific context fields. That operation would be relevant if: A summarization prompt hard-codes one document type, yet the application accepts several document formats. Its corresponding reason or verification is: Named prompt variables allow controlled reuse while retaining explicit instructions for each supported input format. However the problem in this question is: A prompt includes a customer message containing the text ignore previous instructions, and the model treats it as policy. The requirement instead calls for Clearly separate untrusted customer content from trusted system-level instructions; the question-specific check is: Instruction hierarchy and data delimiters reduce the chance that a retrieved or user-supplied string overrides the application's policy.

B:

Require confirmation of transactional prerequisites before making the tool call. A separate source case supports that approach when: A support agent must call a refund tool only after the user confirms an amount. The supporting rationale or observable result is: Workflow-specific sequencing reduces the risk that an agent performs an irreversible action using incomplete user intent. This question describes a different issue: A prompt includes a customer message containing the text ignore previous instructions, and the model treats it as policy. Address it through Clearly separate untrusted customer content from trusted system-level instructions, with the required justification or evidence: Instruction hierarchy and data delimiters reduce the chance that a retrieved or user-supplied string overrides the application's policy.

C:

Define the output schema and constrain responses to the required fields. This is useful for the distinct issue: A customer-service prompt produces long essays when the application expects a three-field JSON object. The relevant technical rationale or evidence for that issue is: Explicit structured output requirements are more reliable than leaving format assumptions to free-form generation. Here the reported issue is: A prompt includes a customer message containing the text ignore previous instructions, and the model treats it as policy. The relevant response is: Clearly separate untrusted customer content from trusted system-level instructions. The first result cannot verify this different correction: Instruction hierarchy and data delimiters reduce the chance that a retrieved or user-supplied string overrides the application's policy.

Question 526

A customer-service prompt produces long essays when the application expects a three-field JSON object. What should be specified?

  1. Require answers to be grounded in provided sources and to abstain when evidence is missing.
  2. Delimit quoted source material and label it as data rather than instructions.
  3. Define the output schema and constrain responses to the required fields.
  4. Parameterize the prompt's document type and task-specific context fields.

Explanation

Correct Answer

C:

Explicit structured output requirements are more reliable than leaving format assumptions to free-form generation.

Incorrect Answers

A:

The alternative, Require answers to be grounded in provided sources and to abstain when evidence is missing, belongs to a different technical condition: A document assistant cites facts not found in the retrieved policy excerpts. There the source-specific mechanism or proof is: Grounding instructions align claims with retrieved context and discourage unsupported details when source material is incomplete. The condition asked about here is: A customer-service prompt produces long essays when the application expects a three-field JSON object. A suitable correction is Define the output schema and constrain responses to the required fields; the targeted justification or result is: Explicit structured output requirements are more reliable than leaving format assumptions to free-form generation.

B:

Delimit quoted source material and label it as data rather than instructions. That operation would be relevant if: A model has to distinguish quoted legal text from the question asked about it, but both appear as unmarked prose. Its corresponding reason or verification is: Clear delimiters reduce confusion between the content to analyze and the instructions controlling analysis. However the problem in this question is: A customer-service prompt produces long essays when the application expects a three-field JSON object. The requirement instead calls for Define the output schema and constrain responses to the required fields; the question-specific check is: Explicit structured output requirements are more reliable than leaving format assumptions to free-form generation.

D:

Parameterize the prompt's document type and task-specific context fields. A separate source case supports that approach when: A summarization prompt hard-codes one document type, yet the application accepts several document formats. The supporting rationale or observable result is: Named prompt variables allow controlled reuse while retaining explicit instructions for each supported input format. This question describes a different issue: A customer-service prompt produces long essays when the application expects a three-field JSON object. Address it through Define the output schema and constrain responses to the required fields, with the required justification or evidence: Explicit structured output requirements are more reliable than leaving format assumptions to free-form generation.

Question 527

A document assistant cites facts not found in the retrieved policy excerpts. What prompt instruction addresses evidence use?

  1. Define the output schema and constrain responses to the required fields.
  2. Require answers to be grounded in provided sources and to abstain when evidence is missing.
  3. Organize prompt sections into clear policy, context, examples, and current task boundaries.
  4. Specify allowed tool actions and validate generated arguments against the approved schema.

Explanation

Correct Answer

B:

Grounding instructions align claims with retrieved context and discourage unsupported details when source material is incomplete.

Incorrect Answers

A:

Define the output schema and constrain responses to the required fields. It addresses another case: A customer-service prompt produces long essays when the application expects a three-field JSON object. For that case, the distinguishing justification or acceptance check is: Explicit structured output requirements are more reliable than leaving format assumptions to free-form generation. The present case instead states: A document assistant cites facts not found in the retrieved policy excerpts. It calls for Require answers to be grounded in provided sources and to abstain when evidence is missing. Showing success for the other case would not establish: Grounding instructions align claims with retrieved context and discourage unsupported details when source material is incomplete.

C:

The alternative, Organize prompt sections into clear policy, context, examples, and current task boundaries, belongs to a different technical condition: An evaluation shows inconsistent outputs because the prompt mixes system policy, retrieved documents, and sample replies into one unstructured block. There the source-specific mechanism or proof is: Role and section structure makes precedence and evidence sources clearer, improving repeatable task execution. The condition asked about here is: A document assistant cites facts not found in the retrieved policy excerpts. A suitable correction is Require answers to be grounded in provided sources and to abstain when evidence is missing; the targeted justification or result is: Grounding instructions align claims with retrieved context and discourage unsupported details when source material is incomplete.

D:

Specify allowed tool actions and validate generated arguments against the approved schema. That operation would be relevant if: A chat application accepts tool arguments created directly from untrusted user text, leading to unsafe operations. Its corresponding reason or verification is: Prompt guidance and downstream validation constrain tool use instead of trusting arbitrary model-generated commands. However the problem in this question is: A document assistant cites facts not found in the retrieved policy excerpts. The requirement instead calls for Require answers to be grounded in provided sources and to abstain when evidence is missing; the question-specific check is: Grounding instructions align claims with retrieved context and discourage unsupported details when source material is incomplete.

Question 528

An agent must return concise answers under a fixed application token budget but often writes paragraphs of unnecessary detail. What should be controlled?

  1. Specify a compact answer format and practical output-length constraints.
  2. Require explicit uncertainty or no-answer behavior when supporting material is unavailable.
  3. Remove credentials from prompts and access them only through protected tool infrastructure.
  4. Specify an explicit source citation output contract tied to retrieved document identifiers.

Explanation

Correct Answer

A:

Response-length guidance reduces token waste while keeping required information present in structured output.

Incorrect Answers

B:

Require explicit uncertainty or no-answer behavior when supporting material is unavailable. This is useful for the distinct issue: A prompt instructs an assistant to invent a reference when source documents are absent. The relevant technical rationale or evidence for that issue is: A clear abstention rule discourages plausible-sounding fabricated citations and unsupported assertions in retrieval-based applications. Here the reported issue is: An agent must return concise answers under a fixed application token budget but often writes paragraphs of unnecessary detail. The relevant response is: Specify a compact answer format and practical output-length constraints. The first result cannot verify this different correction: Response-length guidance reduces token waste while keeping required information present in structured output.

C:

Remove credentials from prompts and access them only through protected tool infrastructure. It addresses another case: An agent must not expose internal tool secrets, but its prompt contains the secrets directly. For that case, the distinguishing justification or acceptance check is: Prompt text can surface in logs or outputs; secret storage must be separated from model-visible instructions. The present case instead states: An agent must return concise answers under a fixed application token budget but often writes paragraphs of unnecessary detail. It calls for Specify a compact answer format and practical output-length constraints. Showing success for the other case would not establish: Response-length guidance reduces token waste while keeping required information present in structured output.

D:

The alternative, Specify an explicit source citation output contract tied to retrieved document identifiers, belongs to a different technical condition: An LLM response must include citations to each retrieved paragraph used, yet the prompt never describes citation placement. There the source-specific mechanism or proof is: A citation contract helps connect each supported claim to available retrieval evidence instead of adding untraceable references. The condition asked about here is: An agent must return concise answers under a fixed application token budget but often writes paragraphs of unnecessary detail. A suitable correction is Specify a compact answer format and practical output-length constraints; the targeted justification or result is: Response-length guidance reduces token waste while keeping required information present in structured output.

Question 529

A prompt needs to extract the same fields from many differently worded invoices. What technique can clarify expected behavior?

  1. Define explicit non-negotiable policy boundaries ahead of optional helpfulness guidance.
  2. Specify allowed tool actions and validate generated arguments against the approved schema.
  3. Parameterize the prompt's document type and task-specific context fields.
  4. Include representative few-shot input-output examples for the extraction task.

Explanation

Correct Answer

D:

Examples demonstrate the intended mapping and format beyond vague natural-language directions alone.

Incorrect Answers

A:

Define explicit non-negotiable policy boundaries ahead of optional helpfulness guidance. A separate source case supports that approach when: A prompt says always be helpful, causing the model to prioritize user preference over required safety rules. The supporting rationale or observable result is: Clear priority reduces conflicts where a broad instruction is incorrectly interpreted as permission to violate controls. This question describes a different issue: A prompt needs to extract the same fields from many differently worded invoices. Address it through Include representative few-shot input-output examples for the extraction task, with the required justification or evidence: Examples demonstrate the intended mapping and format beyond vague natural-language directions alone.

B:

Specify allowed tool actions and validate generated arguments against the approved schema. This is useful for the distinct issue: A chat application accepts tool arguments created directly from untrusted user text, leading to unsafe operations. The relevant technical rationale or evidence for that issue is: Prompt guidance and downstream validation constrain tool use instead of trusting arbitrary model-generated commands. Here the reported issue is: A prompt needs to extract the same fields from many differently worded invoices. The relevant response is: Include representative few-shot input-output examples for the extraction task. The first result cannot verify this different correction: Examples demonstrate the intended mapping and format beyond vague natural-language directions alone.

C:

Parameterize the prompt's document type and task-specific context fields. It addresses another case: A summarization prompt hard-codes one document type, yet the application accepts several document formats. For that case, the distinguishing justification or acceptance check is: Named prompt variables allow controlled reuse while retaining explicit instructions for each supported input format. The present case instead states: A prompt needs to extract the same fields from many differently worded invoices. It calls for Include representative few-shot input-output examples for the extraction task. Showing success for the other case would not establish: Examples demonstrate the intended mapping and format beyond vague natural-language directions alone.

Question 530

A policy summarizer mishandles a rare exception because the prompt demonstrates only normal cases. What should be added?

  1. Ask for essential missing inputs instead of guessing protected identifiers.
  2. Require confirmation of transactional prerequisites before making the tool call.
  3. Include edge-case examples illustrating the policy exception and expected handling.
  4. Require answers to be grounded in provided sources and to abstain when evidence is missing.

Explanation

Correct Answer

C:

Counterexamples help the model distinguish exceptions from common cases that otherwise dominate the prompt instructions.

Incorrect Answers

A:

Ask for essential missing inputs instead of guessing protected identifiers. A separate source case supports that approach when: A model responds to an ambiguous request by fabricating a missing customer identifier rather than asking a question. The supporting rationale or observable result is: Clarification rules reduce unauthorized assumptions and prevent unsafe model actions when critical parameters are absent. This question describes a different issue: A policy summarizer mishandles a rare exception because the prompt demonstrates only normal cases. Address it through Include edge-case examples illustrating the policy exception and expected handling, with the required justification or evidence: Counterexamples help the model distinguish exceptions from common cases that otherwise dominate the prompt instructions.

B:

Require confirmation of transactional prerequisites before making the tool call. This is useful for the distinct issue: A support agent must call a refund tool only after the user confirms an amount. The relevant technical rationale or evidence for that issue is: Workflow-specific sequencing reduces the risk that an agent performs an irreversible action using incomplete user intent. Here the reported issue is: A policy summarizer mishandles a rare exception because the prompt demonstrates only normal cases. The relevant response is: Include edge-case examples illustrating the policy exception and expected handling. The first result cannot verify this different correction: Counterexamples help the model distinguish exceptions from common cases that otherwise dominate the prompt instructions.

D:

Require answers to be grounded in provided sources and to abstain when evidence is missing. It addresses another case: A document assistant cites facts not found in the retrieved policy excerpts. For that case, the distinguishing justification or acceptance check is: Grounding instructions align claims with retrieved context and discourage unsupported details when source material is incomplete. The present case instead states: A policy summarizer mishandles a rare exception because the prompt demonstrates only normal cases. It calls for Include edge-case examples illustrating the policy exception and expected handling. Showing success for the other case would not establish: Counterexamples help the model distinguish exceptions from common cases that otherwise dominate the prompt instructions.

Operationalizing Machine Learning and Generative AI Solutions (AI-300) Practice Tests