Practice Exams:

Microsoft AI-103 Dumps

EXAM: Developing AI Apps and Agents on Azure
Verified By Experts
Premium Bundle
$49.99

AI-103 Premium ETE File


  • Premium File 187 Questions & Answers
Whats Included:
  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates

Microsoft AI-103 Practice Test 61

Question 600

During a support escalation, a machine-learning developer must accept a change for parallel tool retrieval. The issue to be addressed is: An agent waits sequentially for independent lookup tools. Which concrete verification action demonstrates that the relevant behavior works?

  1. Stress the endpoint and inspect retry rates and eventual failures
  2. Compare end-to-end latency for sequential and concurrent read cases
  3. Parallelize dependent write operations without ordering or authorization checks
  4. Parallelize independent safe read-only tool calls with bounded concurrency

Explanation

Correct Answer

B:

To verify this change, the education technology company should compare end-to-end latency for sequential and concurrent read cases. Operators still need the implementation to parallelize independent safe read-only tool calls with bounded concurrency and the preventive requirement to limit parallel calls and test cancellation across branches.

Incorrect Answers

A:

For retry backoff, it can be sensible to stress the endpoint and inspect retry rates and eventual failures if a bursty application receives 429 responses from an inference endpoint. Without the actual change to compare end-to-end latency for sequential and concurrent read cases, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.

C:

In the education technology company acceptance review, trying to parallelize dependent write operations without ordering or authorization checks does not test the exact parallel tool retrieval condition where an agent waits sequentially for independent lookup tools; the relevant test is to compare end-to-end latency for sequential and concurrent read cases. Acceptance should instead demonstrate that the service can compare end-to-end latency for sequential and concurrent read cases, using the described failure as the test criterion.

D:

The action to parallelize independent safe read-only tool calls with bounded concurrency can contribute to parallel tool retrieval as a implementation step; here, an agent waits sequentially for independent lookup tools. In the education technology company acceptance review, trying to parallelize independent safe read-only tool calls with bounded concurrency does not test the exact parallel tool retrieval condition where an agent waits sequentially for independent lookup tools; the relevant test is to compare end-to-end latency for sequential and concurrent read cases.

Question 601

At a public library network, the AI application owner is preparing a staged service rollout. The concrete requirement is: An AI app passes common questions but fails rare high-risk intents. Which design action BEST addresses the stated need?

  1. Inspect error rates across normal and high-risk test categories
  2. Summarize only authorized essential state into a compact durable record
  3. Stratify evaluation cases by intent difficulty and harm severity
  4. Report one aggregate accuracy number without examining critical failure cohorts

Explanation

Correct Answer

C:

The proposed implementation is to stratify evaluation cases by intent difficulty and harm severity; this matches the operational requirement that an AI app passes common questions but fails rare high-risk intents. The team can inspect error rates across normal and high-risk test categories. The team should also set separate safety thresholds for rare high-impact tasks.

Incorrect Answers

A:

The action to inspect error rates across normal and high-risk test categories can contribute to evaluation stratification as a targeted acceptance check step; here, an AI app passes common questions but fails rare high-risk intents. Choosing to inspect error rates across normal and high-risk test categories cannot directly replace the specified action to stratify evaluation cases by intent difficulty and harm severity when an AI app passes common questions but fails rare high-risk intents.

B:

Choosing to summarize only authorized essential state into a compact durable record would direct effort toward context compression, whose characteristic issue is that a long agent session has excessive token cost and truncates relevant history. The relevant corrective or preventive action here remains to stratify evaluation cases by intent difficulty and harm severity, with its own validation criteria.

D:

Choosing to report one aggregate accuracy number without examining critical failure cohorts cannot directly replace the specified action to stratify evaluation cases by intent difficulty and harm severity when an AI app passes common questions but fails rare high-risk intents. The actual implementation must stratify evaluation cases by intent difficulty and harm severity, because the stated service condition is that an AI app passes common questions but fails rare high-risk intents.

Question 602

After a release acceptance review at a software consultancy, the team observes this failure: An AI app passes common questions but fails rare high-risk intents. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?

  1. Stratify evaluation cases by intent difficulty and harm severity
  2. Report one aggregate accuracy number without examining critical failure cohorts
  3. Assume natural-language prompt edits cannot alter factual answer behavior
  4. Inspect error rates across normal and high-risk test categories

Explanation

Correct Answer

B:

This design mistake is to report one aggregate accuracy number without examining critical failure cohorts; it conflicts with the actual evaluation stratification behavior needed when an AI app passes common questions but fails rare high-risk intents. The software consultancy should instead stratify evaluation cases by intent difficulty and harm severity, followed by a check to inspect error rates across normal and high-risk test categories.

Incorrect Answers

A:

The alternative to stratify evaluation cases by intent difficulty and harm severity is a implementation step rather than the faulty assumption being diagnosed. This software consultancy failure-analysis alternative of stratify evaluation cases by intent difficulty and harm severity does not diagnose the particular evaluation stratification defect where an AI app passes common questions but fails rare high-risk intents; the keyed mistake is report one aggregate accuracy number without examining critical failure cohorts.

C:

The option to assume natural-language prompt edits cannot alter factual answer behavior would normally serve a different prompt version rollback requirement, where a revised system prompt changes correct responses on known tasks. The alternative may be useful in its own context, but it does not directly supply report one aggregate accuracy number without examining critical failure cohorts for the stated decision.

D:

The alternative to inspect error rates across normal and high-risk test categories is a targeted acceptance check step rather than the faulty assumption being diagnosed. This software consultancy failure-analysis alternative of inspect error rates across normal and high-risk test categories does not diagnose the particular evaluation stratification defect where an AI app passes common questions but fails rare high-risk intents; the keyed mistake is report one aggregate accuracy number without examining critical failure cohorts.

Question 603

The regional hospital is responding to a design meeting. A AI platform engineer discovers this issue: An AI app passes common questions but fails rare high-risk intents. What corrective step should the team take to restore the intended functionality most directly?

  1. Report one aggregate accuracy number without examining critical failure cohorts
  2. Add representative high-risk tests and repair specific failures
  3. Inspect error rates across normal and high-risk test categories
  4. Reduce unnecessary completion length and compare task-quality loss

Explanation

Correct Answer

B:

The direct repair is to add representative high-risk tests and repair specific failures. Once the condition is corrected, the regional hospital should inspect error rates across normal and high-risk test categories. To reduce recurrence, it should also set separate safety thresholds for rare high-impact tasks; those steps provide different recovery and prevention functions.

Incorrect Answers

A:

During regional hospital remediation, the proposed step to report one aggregate accuracy number without examining critical failure cohorts is not the direct restoration action for evaluation stratification when an AI app passes common questions but fails rare high-risk intents; the keyed repair is add representative high-risk tests and repair specific failures. After remediation the team must inspect error rates across normal and high-risk test categories; the action in this option does not itself establish that target recovery outcome.

C:

The action to inspect error rates across normal and high-risk test categories can contribute to evaluation stratification as a targeted acceptance check step; here, an AI app passes common questions but fails rare high-risk intents. During regional hospital remediation, the proposed step to inspect error rates across normal and high-risk test categories is not the direct restoration action for evaluation stratification when an AI app passes common questions but fails rare high-risk intents; the keyed repair is add representative high-risk tests and repair specific failures.

D:

This approach relates to latency under long outputs: a support bot's response time increases with unbounded verbose generation. If the team chose it instead, the targeted check to inspect error rates across normal and high-risk test categories could still fail because the underlying required action is to add representative high-risk tests and repair specific failures.

Question 604

A compliance specialist at a insurance platform wants the following issue to be prevented in later deployments: An AI app passes common questions but fails rare high-risk intents. Which enduring control or engineering guardrail BEST prevents recurrence?

  1. Set separate safety thresholds for rare high-impact tasks
  2. Add representative high-risk tests and repair specific failures
  3. Require prompt version hashes and an evaluation gate
  4. Report one aggregate accuracy number without examining critical failure cohorts

Explanation

Correct Answer

A:

Standardizing the boundary to set separate safety thresholds for rare high-impact tasks is preventive because the same evaluation stratification failure could otherwise return: an AI app passes common questions but fails rare high-risk intents. For this insurance platform, the service should also stratify evaluation cases by intent difficulty and harm severity, and tests should inspect error rates across normal and high-risk test categories so that the ongoing requirement is measurable.

Incorrect Answers

B:

The action to add representative high-risk tests and repair specific failures can contribute to evaluation stratification as a corrective remediation step; here, an AI app passes common questions but fails rare high-risk intents. For prevention at the insurance platform, the measure to add representative high-risk tests and repair specific failures is weaker or aimed at another issue than the evaluation stratification guardrail to set separate safety thresholds for rare high-impact tasks needed when an AI app passes common questions but fails rare high-risk intents.

C:

The action to require prompt version hashes and an evaluation gate addresses prompt version rollback rather than the decisive evaluation stratification condition. The insurance platform should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to set separate safety thresholds for rare high-impact tasks.

D:

For prevention at the insurance platform, the measure to report one aggregate accuracy number without examining critical failure cohorts is weaker or aimed at another issue than the evaluation stratification guardrail to set separate safety thresholds for rare high-impact tasks needed when an AI app passes common questions but fails rare high-risk intents. The lasting guardrail for this specific threat is to set separate safety thresholds for rare high-impact tasks, not simply to complete an adjacent service task.

Question 605

During a release acceptance review, a product support lead must accept a change for evaluation stratification. The issue to be addressed is: An AI app passes common questions but fails rare high-risk intents. Which concrete verification action demonstrates that the relevant behavior works?

  1. Report one aggregate accuracy number without examining critical failure cohorts
  2. Inspect error rates across normal and high-risk test categories
  3. Stratify evaluation cases by intent difficulty and harm severity
  4. Set separate safety thresholds for rare high-impact tasks
  5. Ask questions absent from the knowledge store and verify uncertainty

Explanation

Correct Answer

B:

The targeted acceptance test is to inspect error rates across normal and high-risk test categories. The design should first stratify evaluation cases by intent difficulty and harm severity. A robust media platform deployment also needs to set separate safety thresholds for rare high-impact tasks; passing this check provides relevant evidence but does not certify every possible input or security risk.

Incorrect Answers

A:

In the media platform acceptance review, trying to report one aggregate accuracy number without examining critical failure cohorts does not test the exact evaluation stratification condition where an AI app passes common questions but fails rare high-risk intents; the relevant test is to inspect error rates across normal and high-risk test categories. Acceptance should instead demonstrate that the service can inspect error rates across normal and high-risk test categories, using the described failure as the test criterion.

C:

The action to stratify evaluation cases by intent difficulty and harm severity can contribute to evaluation stratification as a implementation step; here, an AI app passes common questions but fails rare high-risk intents. In the media platform acceptance review, trying to stratify evaluation cases by intent difficulty and harm severity does not test the exact evaluation stratification condition where an AI app passes common questions but fails rare high-risk intents; the relevant test is to inspect error rates across normal and high-risk test categories.

D:

The action to set separate safety thresholds for rare high-impact tasks can contribute to evaluation stratification as a ongoing preventive rule step; here, an AI app passes common questions but fails rare high-risk intents. In the media platform acceptance review, trying to set separate safety thresholds for rare high-impact tasks does not test the exact evaluation stratification condition where an AI app passes common questions but fails rare high-risk intents; the relevant test is to inspect error rates across normal and high-risk test categories.

E:

Choosing to ask questions absent from the knowledge store and verify uncertainty would direct effort toward hallucination counterexample tests, whose characteristic issue is that a new prompt handles known facts but invents missing ones. The relevant corrective or preventive action here remains to inspect error rates across normal and high-risk test categories, with its own validation criteria.

Question 606

At a education technology company, the data engineer is preparing a quality regression test. The concrete requirement is: A new prompt handles known facts but invents missing ones. Which design action BEST addresses the stated need?

  1. Ask questions absent from the knowledge store and verify uncertainty
  2. Reward every generated answer even when no evidence exists
  3. Tag each request trace with model deployment prompt version and tool call IDs
  4. Include intentionally unanswerable cases and score abstention appropriately

Explanation

Correct Answer

D:

Choosing to include intentionally unanswerable cases and score abstention appropriately directly targets the described hallucination counterexample tests gap: a new prompt handles known facts but invents missing ones. As a lasting boundary, they also need to build eval sets with supported and unsupported claims, because an implementation by itself does not prove safe operation.

Incorrect Answers

A:

The action to ask questions absent from the knowledge store and verify uncertainty can contribute to hallucination counterexample tests as a targeted acceptance check step; here, a new prompt handles known facts but invents missing ones. Choosing to ask questions absent from the knowledge store and verify uncertainty cannot directly replace the specified action to include intentionally unanswerable cases and score abstention appropriately when a new prompt handles known facts but invents missing ones.

B:

Choosing to reward every generated answer even when no evidence exists cannot directly replace the specified action to include intentionally unanswerable cases and score abstention appropriately when a new prompt handles known facts but invents missing ones. The actual implementation must include intentionally unanswerable cases and score abstention appropriately, because the stated service condition is that a new prompt handles known facts but invents missing ones.

C:

Choosing to tag each request trace with model deployment prompt version and tool call IDs would direct effort toward production trace correlation, whose characteristic issue is that a user complaint cannot be tied to a specific model and prompt deployment. The relevant corrective or preventive action here remains to include intentionally unanswerable cases and score abstention appropriately, with its own validation criteria.

Question 607

After a production incident at a pharmaceutical research group, the team observes this failure: A new prompt handles known facts but invents missing ones. Which flawed assumption or implementation choice would MOST DIRECTLY explain why the intended outcome was not achieved?

  1. Ask questions absent from the knowledge store and verify uncertainty
  2. Reward every generated answer even when no evidence exists
  3. Include intentionally unanswerable cases and score abstention appropriately
  4. Cache every response indefinitely without version awareness

Explanation

Correct Answer

B:

The faulty assumption is reward every generated answer even when no evidence exists. A functioning alternative would be to include intentionally unanswerable cases and score abstention appropriately, and the relevant validation would be to ask questions absent from the knowledge store and verify uncertainty.

Incorrect Answers

A:

The alternative to ask questions absent from the knowledge store and verify uncertainty is a targeted acceptance check step rather than the faulty assumption being diagnosed. This pharmaceutical research group failure-analysis alternative of ask questions absent from the knowledge store and verify uncertainty does not diagnose the particular hallucination counterexample tests defect where a new prompt handles known facts but invents missing ones; the keyed mistake is reward every generated answer even when no evidence exists.

C:

The alternative to include intentionally unanswerable cases and score abstention appropriately is a implementation step rather than the faulty assumption being diagnosed. This pharmaceutical research group failure-analysis alternative of include intentionally unanswerable cases and score abstention appropriately does not diagnose the particular hallucination counterexample tests defect where a new prompt handles known facts but invents missing ones; the keyed mistake is reward every generated answer even when no evidence exists.

D:

This pharmaceutical research group failure-analysis alternative of cache every response indefinitely without version awareness does not diagnose the particular hallucination counterexample tests defect where a new prompt handles known facts but invents missing ones; the keyed mistake is reward every generated answer even when no evidence exists. The alternative may be useful in its own context, but it does not directly supply reward every generated answer even when no evidence exists for the stated decision.

Question 608

The travel support provider is responding to a data protection inspection. A cloud developer discovers this issue: A new prompt handles known facts but invents missing ones. What corrective step should the team take to restore the intended functionality most directly?

  1. Reward every generated answer even when no evidence exists
  2. Ask questions absent from the knowledge store and verify uncertainty
  3. Restore tool-boundary restrictions and replay adversarial cases
  4. Tighten source-grounding conditions and retest missing-fact cases

Explanation

Correct Answer

D:

The appropriate remediation is to tighten source-grounding conditions and retest missing-fact cases, because the failure concerns hallucination counterexample tests: a new prompt handles known facts but invents missing ones. A meaningful verification should ask questions absent from the knowledge store and verify uncertainty, and the durable operating rule should build eval sets with supported and unsupported claims after the immediate fix.

Incorrect Answers

A:

During travel support provider remediation, the proposed step to reward every generated answer even when no evidence exists is not the direct restoration action for hallucination counterexample tests when a new prompt handles known facts but invents missing ones; the keyed repair is tighten source-grounding conditions and retest missing-fact cases. After remediation the team must ask questions absent from the knowledge store and verify uncertainty; the action in this option does not itself establish that target recovery outcome.

B:

The action to ask questions absent from the knowledge store and verify uncertainty can contribute to hallucination counterexample tests as a targeted acceptance check step; here, a new prompt handles known facts but invents missing ones. During travel support provider remediation, the proposed step to ask questions absent from the knowledge store and verify uncertainty is not the direct restoration action for hallucination counterexample tests when a new prompt handles known facts but invents missing ones; the keyed repair is tighten source-grounding conditions and retest missing-fact cases.

C:

During travel support provider remediation, the proposed step to restore tool-boundary restrictions and replay adversarial cases is not the direct restoration action for hallucination counterexample tests when a new prompt handles known facts but invents missing ones; the keyed repair is tighten source-grounding conditions and retest missing-fact cases. If the team chose it instead, the targeted check to ask questions absent from the knowledge store and verify uncertainty could still fail because the underlying required action is to tighten source-grounding conditions and retest missing-fact cases.

Question 609

A site reliability engineer at a scientific publisher wants the following issue to be prevented in later deployments: A new prompt handles known facts but invents missing ones. Which enduring control or engineering guardrail BEST prevents recurrence?

  1. Tighten source-grounding conditions and retest missing-fact cases
  2. Reward every generated answer even when no evidence exists
  3. Build eval sets with supported and unsupported claims
  4. Monitor tokens and evidence recall for retrieval settings

Explanation

Correct Answer

C:

The sustained prevention measure is to build eval sets with supported and unsupported claims. The implementation can still use the separate action to include intentionally unanswerable cases and score abstention appropriately, but a guardrail constrains future changes. The scientific publisher can measure whether the rule works by arranging to ask questions absent from the knowledge store and verify uncertainty.

Incorrect Answers

A:

The action to tighten source-grounding conditions and retest missing-fact cases can contribute to hallucination counterexample tests as a corrective remediation step; here, a new prompt handles known facts but invents missing ones. For prevention at the scientific publisher, the measure to tighten source-grounding conditions and retest missing-fact cases is weaker or aimed at another issue than the hallucination counterexample tests guardrail to build eval sets with supported and unsupported claims needed when a new prompt handles known facts but invents missing ones.

B:

For prevention at the scientific publisher, the measure to reward every generated answer even when no evidence exists is weaker or aimed at another issue than the hallucination counterexample tests guardrail to build eval sets with supported and unsupported claims needed when a new prompt handles known facts but invents missing ones. The lasting guardrail for this specific threat is to build eval sets with supported and unsupported claims, not simply to complete an adjacent service task.

D:

The action to monitor tokens and evidence recall for retrieval settings addresses retrieved context trimming rather than the decisive hallucination counterexample tests condition. The scientific publisher should not infer that fixing one adjacent AI service validates its unrelated workflow; the requested action is to build eval sets with supported and unsupported claims.

Question 610

During a production incident, a security engineer must accept a change for hallucination counterexample tests. The issue to be addressed is: A new prompt handles known facts but invents missing ones. Which concrete verification action demonstrates that the relevant behavior works?

  1. Include intentionally unanswerable cases and score abstention appropriately
  2. Ask questions absent from the knowledge store and verify uncertainty
  3. Reward every generated answer even when no evidence exists
  4. Construct a false shared-premise case and verify it is rejected

Explanation

Correct Answer

B:

To verify this change, the nonprofit organization should ask questions absent from the knowledge store and verify uncertainty. Operators still need the implementation to include intentionally unanswerable cases and score abstention appropriately and the preventive requirement to build eval sets with supported and unsupported claims.

Incorrect Answers

A:

The action to include intentionally unanswerable cases and score abstention appropriately can contribute to hallucination counterexample tests as a implementation step; here, a new prompt handles known facts but invents missing ones. In the nonprofit organization acceptance review, trying to include intentionally unanswerable cases and score abstention appropriately does not test the exact hallucination counterexample tests condition where a new prompt handles known facts but invents missing ones; the relevant test is to ask questions absent from the knowledge store and verify uncertainty.

C:

In the nonprofit organization acceptance review, trying to reward every generated answer even when no evidence exists does not test the exact hallucination counterexample tests condition where a new prompt handles known facts but invents missing ones; the relevant test is to ask questions absent from the knowledge store and verify uncertainty. Acceptance should instead demonstrate that the service can ask questions absent from the knowledge store and verify uncertainty, using the described failure as the test criterion.

D:

For multimodel consensus risk, it can be sensible to construct a false shared-premise case and verify it is rejected if two models agree on an answer but share the same unsupported source. Without the actual change to ask questions absent from the knowledge store and verify uncertainty, the answer lacks the correct operational scope even if the alternate action is technically valid elsewhere.

Developing AI Apps and Agents on Azure (AI-103) Practice Tests