McAfee Secure

Exam Code: AI-901

Exam Name: Microsoft Azure AI Fundamentals

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Azure AI Fundamentals

Microsoft AI-901 Questions & Answers

Study with Up-To-Date REAL Exam Questions and Answers from the ACTUAL Test

166 Questions & Answers with Testing Engine

"Microsoft Azure AI Fundamentals Exam", also known as AI-901 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-901 Exam Engine. Your AI-901 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-901 Practice Test 76

Question 750

An Azure AI Fundamentals solution has this requirement: A developer wants to compare prompt revisions objectively. Which implementation most directly addresses it?

  1. Limit retrieval and chat context to the authenticated case scope
  2. Use a versioned evaluation set with stable scoring and test inputs
  3. Write an instruction to identify missing inputs and request clarification before tools
  4. Send the image in a supported multimodal message along with the question

Explanation

Correct Answer

B:

Anecdotal success on different prompts cannot establish an improvement. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Source scoping enforces privacy and also improves answer relevance. That is not sufficient here: Anecdotal success on different prompts cannot establish an improvement.

C:

Prompts can define required preconditions but tools must still enforce them. That is not sufficient here: Anecdotal success on different prompts cannot establish an improvement.

D:

The model must receive the actual visual content to reason about it. That is not sufficient here: Anecdotal success on different prompts cannot establish an improvement.

Question 751

Troubleshooting report: The system submits the entire oversized document on every call. Which correction is MOST directly targeted to this failure?

  1. Chunk and retrieve relevant source sections before inference
  2. Calculate peak token and request demand and request a supported quota or capacity change
  3. Verify claims against trusted evidence and permit abstention
  4. Implement and test fixed numeric rules in application code

Explanation

Correct Answer

A:

This correction directly addresses the reported failure. A finite context window limits how much material can be processed at once. The targeted change is: Chunk and retrieve relevant source sections before inference.

Incorrect Answers

B:

Quota sizing is a deployment-capacity decision; retrying a predictable sustained excess cannot add model throughput. In the stated case, troubleshooting report: The system submits the entire oversized document on every call; “Calculate peak token and request demand and request a supported quota or capacity change” does not target that failure.

C:

Language-model likelihood does not independently establish factual truth. In the stated case, troubleshooting report: The system submits the entire oversized document on every call; “Verify claims against trusted evidence and permit abstention” does not target that failure.

D:

Generative models are not a substitute for auditable deterministic computation. In the stated case, troubleshooting report: The system submits the entire oversized document on every call; “Implement and test fixed numeric rules in application code” does not target that failure.

Question 752

An Azure AI solution uses the following approved control: Confirm the model deployment identifier and endpoint resource match. What observable result should pass QA?

  1. A permitted larger output budget allows the complete response
  2. The client succeeds using the deployment configured on the intended resource
  3. A regression test records model version and key generation settings
  4. A held-out specialty test improves over prompting and retrieval baselines

Explanation

Correct Answer

B:

Applications call a deployed model resource rather than a generic marketing model name. The expected acceptance evidence is: The client succeeds using the deployment configured on the intended resource. This observation directly tests the stated behavior.

Incorrect Answers

A:

Generation stops when an output budget or stop condition is reached. This observation does not establish the required result: The client succeeds using the deployment configured on the intended resource.

C:

Reproducibility depends on more than the visible prompt text. This observation does not establish the required result: The client succeeds using the deployment configured on the intended resource.

D:

Model specialization may help task behavior but requires measured benefit. This observation does not establish the required result: The client succeeds using the deployment configured on the intended resource.

Question 753

An AI operations team is testing two distinct capabilities. Failure (1): The team assumes a model recalls every earlier session without storage. Failure (2): A low max-output-token configuration cuts off the answer. Which TWO specific corrections should the team apply?

  1. Demonstrate probabilistic token generation with contrasting prompts
  2. Run task-specific evaluations before choosing deployment size
  3. Increase the supported limit within deployment quotas or shorten the requested output
  4. Separate behavioral model errors from stale or absent knowledge
  5. Provide required context or a governed retrieval/memory mechanism

Explanation

Correct Answers

C:

Correct for failure (2). Generation stops when an output budget or stop condition is reached. It independently corrects the second defect.

E:

Correct for failure (1). A model conditions its output on supplied instructions and conversation text. The proposed correction directly addresses that reported defect.

Incorrect Answers

A:

A generative language model estimates likely continuations rather than reading a stored answer paragraph. The stated pair of failures instead requires “Increase the supported limit within deployment quotas or shorten the requested output” and “Provide required context or a governed retrieval/memory mechanism”; this option cannot substitute for both.

B:

Model size alone does not establish suitability for a particular task. The stated pair of failures instead requires “Increase the supported limit within deployment quotas or shorten the requested output” and “Provide required context or a governed retrieval/memory mechanism”; this option cannot substitute for both.

D:

Model specialization may help task behavior but requires measured benefit. The stated pair of failures instead requires “Increase the supported limit within deployment quotas or shorten the requested output” and “Provide required context or a governed retrieval/memory mechanism”; this option cannot substitute for both.

Question 754

Troubleshooting report: A downstream parser accepts arbitrary surrounding prose as valid JSON. Which correction is MOST directly targeted to this failure?

  1. Tighten the output contract and apply post-response validation
  2. Enforce a typed response schema and fail invalid output
  3. Capture per-request usage and attribute it to versioned deployments
  4. Add idempotency keys and query the original transaction state

Explanation

Correct Answer

B:

This correction directly addresses the reported failure. Model output constraints help downstream parsing but still require validation. The targeted change is: Enforce a typed response schema and fail invalid output.

Incorrect Answers

A:

Explicit examples can improve adherence to bounded output conventions. In the stated case, troubleshooting report: A downstream parser accepts arbitrary surrounding prose as valid JSON; “Tighten the output contract and apply post-response validation” does not target that failure.

C:

Usage telemetry reveals context-growth and unexpected repetition. In the stated case, troubleshooting report: A downstream parser accepts arbitrary surrounding prose as valid JSON; “Capture per-request usage and attribute it to versioned deployments” does not target that failure.

D:

A timeout does not prove that the original mutation failed. In the stated case, troubleshooting report: A downstream parser accepts arbitrary surrounding prose as valid JSON; “Add idempotency keys and query the original transaction state” does not target that failure.

Question 755

Troubleshooting report: The team treats the task as selecting one of only two predefined classes. Which correction is MOST directly targeted to this failure?

  1. Analyze multiple temporal segments and preserve timecodes
  2. Select a generative response workflow instead of a binary classifier
  3. Use classification outputs for the entire image
  4. Treat absent fields as unknown and request a better image

Explanation

Correct Answer

B:

This correction directly addresses the reported failure. Generation produces novel text, unlike choosing a fixed class label. The targeted change is: Select a generative response workflow instead of a binary classifier.

Incorrect Answers

A:

Video understanding retains order and time, unlike single-image tagging. In the stated case, troubleshooting report: The team treats the task as selecting one of only two predefined classes; “Analyze multiple temporal segments and preserve timecodes” does not target that failure.

C:

Image classification predicts image-level labels. In the stated case, troubleshooting report: The team treats the task as selecting one of only two predefined classes; “Use classification outputs for the entire image” does not target that failure.

D:

Images of receipts need OCR/layout extraction before typed business values. In the stated case, troubleshooting report: The team treats the task as selecting one of only two predefined classes; “Treat absent fields as unknown and request a better image” does not target that failure.

Question 756

An acceptance tester needs to validate: Include a verifiable source link and section reference alongside material policy answers. Which observed result proves the intended behavior?

  1. Each policy answer citation resolves to the correct approved paragraph
  2. A screen-reader user can complete the service flow without mouse-only actions
  3. The transfer remains pending until delegated approval is recorded
  4. Eligible memory entries disappear from active stores after the verified request

Explanation

Correct Answer

A:

Transparency lets users inspect evidence instead of accepting fluent text. The expected acceptance evidence is: Each policy answer citation resolves to the correct approved paragraph. This observation directly tests the stated behavior.

Incorrect Answers

B:

Inclusive design permits nonvisual navigation of the same task path. This observation does not establish the required result: Each policy answer citation resolves to the correct approved paragraph.

C:

Accountable high-impact operations need verifiable authorized signatures. This observation does not establish the required result: Each policy answer citation resolves to the correct approved paragraph.

D:

Accountability requires evidence of actual data lifecycle actions. This observation does not establish the required result: Each policy answer citation resolves to the correct approved paragraph.

Question 757

An acceptance tester needs to validate: Distinguish generation variability controls from evidence-based factual accuracy. Which observed result proves the intended behavior?

  1. Known unsupported claims remain flagged regardless of temperature
  2. A follow-up pronoun resolves to the correct earlier referent
  3. A trace shows which tool result supported each claimed business action
  4. A proposed snippet passes syntax and behavioral tests before merge

Explanation

Correct Answer

A:

Sampling parameters influence diversity but do not verify knowledge. The expected acceptance evidence is: Known unsupported claims remain flagged regardless of temperature. This observation directly tests the stated behavior.

Incorrect Answers

B:

Stateful conversations require supplying or retrieving relevant prior context. This observation does not establish the required result: Known unsupported claims remain flagged regardless of temperature.

C:

Operational observability aids debugging without exposing hidden sensitive prompt data. This observation does not establish the required result: Known unsupported claims remain flagged regardless of temperature.

D:

Generated code must be validated, not treated as production-ready from fluency. This observation does not establish the required result: Known unsupported claims remain flagged regardless of temperature.

Question 758

Troubleshooting report: The tool searches unrelated private repositories by default. Which correction is MOST directly targeted to this failure?

  1. Prune irrelevant history and preserve the minimum required state
  2. Keep policy instructions separate from lower-trust data
  3. Restrict the tool to approved documents and identities
  4. Add idempotency keys and query the original transaction state

Explanation

Correct Answer

C:

This correction directly addresses the reported failure. Tool descriptions guide selection, while backend rules enforce allowed sources. The targeted change is: Restrict the tool to approved documents and identities.

Incorrect Answers

A:

Long irrelevant context consumes tokens and may increase latency. In the stated case, troubleshooting report: The tool searches unrelated private repositories by default; “Prune irrelevant history and preserve the minimum required state” does not target that failure.

B:

High-priority instructions should describe approved behavior and escalation rules. In the stated case, troubleshooting report: The tool searches unrelated private repositories by default; “Keep policy instructions separate from lower-trust data” does not target that failure.

D:

A timeout does not prove that the original mutation failed. In the stated case, troubleshooting report: The tool searches unrelated private repositories by default; “Add idempotency keys and query the original transaction state” does not target that failure.

Question 759

An Azure AI solution uses the following approved control: Extract currency code and monetary amount as separate validated fields. What observable result should pass QA?

  1. The output identifies both amount and confirmed currency
  2. Each known document type reaches the expected analyzer
  3. A reply links its claim to a transcript time or relevant video segment
  4. Extracted line items reconcile to visible row amounts

Explanation

Correct Answer

A:

A currency symbol or amount alone may not establish financial meaning. The expected acceptance evidence is: The output identifies both amount and confirmed currency. This observation directly tests the stated behavior.

Incorrect Answers

B:

Different document kinds have distinct field schemas and review paths. This observation does not establish the required result: The output identifies both amount and confirmed currency.

C:

Multimodal grounding should preserve how the answer maps to source media. This observation does not establish the required result: The output identifies both amount and confirmed currency.

D:

Cell relationships are necessary to check item totals. This observation does not establish the required result: The output identifies both amount and confirmed currency.

Question 760

Troubleshooting report: The pipeline performs only speech recognition and produces no audio reply. Which correction is MOST directly targeted to this failure?

  1. Add realistic noise tests and a clarification fallback
  2. Add an authorized synthesis stage for the answer
  3. Supply an authorized identity and matching resource configuration
  4. Add a no-speech branch and clarification option

Explanation

Correct Answer

B:

This correction directly addresses the reported failure. Input transcription and output synthesis are distinct steps. The targeted change is: Add an authorized synthesis stage for the answer.

Incorrect Answers

A:

Signal quality directly affects recognizer error rates. In the stated case, troubleshooting report: The pipeline performs only speech recognition and produces no audio reply; “Add realistic noise tests and a clarification fallback” does not target that failure.

C:

Audio capture success does not establish remote service authentication. In the stated case, troubleshooting report: The pipeline performs only speech recognition and produces no audio reply; “Supply an authorized identity and matching resource configuration” does not target that failure.

D:

An empty recognition result is not semantically equivalent to the word no. In the stated case, troubleshooting report: The pipeline performs only speech recognition and produces no audio reply; “Add a no-speech branch and clarification option” does not target that failure.

Microsoft Azure AI Fundamentals (AI-901) Practice Tests