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 56

Question 550

An AI operations team is testing two distinct capabilities. Failure (1): The checkbox is discarded during plain text extraction. Failure (2): The model fabricates a purchase-order number. Which TWO specific corrections should the team apply?

  1. Keep transcript anchors and validate extracted tasks
  2. Include selection-mark processing in the analyzer
  3. Use table-aware layout extraction and numeric validation
  4. Enable supported layout and field extraction features
  5. Validate field presence against source evidence

Explanation

Correct Answers

B:

Correct for failure (1). Checkbox state is a structured signal rather than arbitrary prose. The proposed correction directly addresses that reported defect.

E:

Correct for failure (2). Extraction should preserve absence and distinguish it from zero. It independently corrects the second defect.

Incorrect Answers

A:

Transcription and summarization are separate results with different validation. The stated pair of failures instead requires “Include selection-mark processing in the analyzer” and “Validate field presence against source evidence”; this option cannot substitute for both.

C:

Cell relationships are necessary to check item totals. It does not complete the two decisions required in this question.

D:

Ordinary plain-text OCR may lose table relationships and checkbox states. The stated pair of failures instead requires “Include selection-mark processing in the analyzer” and “Validate field presence against source evidence”; this option cannot substitute for both.

Question 551

A solution team has selected this control: Apply approved audio retention and access rules and keep only required derived text. Which observation is the most direct acceptance evidence?

  1. The output audio reads the approved sentence intelligibly
  2. A test email returns correct product and organization spans
  3. Raw call recordings expire while authorized transcripts remain per policy
  4. A summary preserves issue, action and resolution from the ticket

Explanation

Correct Answer

C:

Speech workloads still process personal data and require controlled lifecycle. The expected acceptance evidence is: Raw call recordings expire while authorized transcripts remain per policy. This observation directly tests the stated behavior.

Incorrect Answers

A:

Speech synthesis generates audio waveforms from supplied text. This observation does not establish the required result: Raw call recordings expire while authorized transcripts remain per policy.

B:

NER identifies named spans rather than summarizing the email. This observation does not establish the required result: Raw call recordings expire while authorized transcripts remain per policy.

D:

Summarization compresses written information rather than changing its facts. This observation does not establish the required result: Raw call recordings expire while authorized transcripts remain per policy.

Question 552

Troubleshooting report: The application deploys generated code immediately without review. Which correction is MOST directly targeted to this failure?

  1. Isolate request state and bind responses to their originating session
  2. Constrain the call to the documented JSON input schema
  3. Keep publication behind a separate approved deployment step
  4. Use code validation and protected deployment controls

Explanation

Correct Answer

D:

This correction directly addresses the reported failure. Generated code must be validated, not treated as production-ready from fluency. The targeted change is: Use code validation and protected deployment controls.

Incorrect Answers

A:

Concurrent inference must not mix user data or response ownership. In the stated case, troubleshooting report: The application deploys generated code immediately without review; “Isolate request state and bind responses to their originating session” does not target that failure.

B:

Invalid tool parameters break calls even if the intended operation is correct. In the stated case, troubleshooting report: The application deploys generated code immediately without review; “Constrain the call to the documented JSON input schema” does not target that failure.

C:

A draft or tested agent is not automatically a publicly released service. In the stated case, troubleshooting report: The application deploys generated code immediately without review; “Keep publication behind a separate approved deployment step” does not target that failure.

Question 553

An Azure-based AI team must satisfy this objective: A team deploys a model but forgot to test supported input token limits. Which design choice is best?

  1. Compare model deployment, prompt version, retrieval index and tool configuration
  2. Store the approved instruction as versioned application configuration
  3. Check supported context and output limits for the chosen deployment
  4. Place the current request in the supported user message field

Explanation

Correct Answer

C:

Capabilities and limits vary by model deployment. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Environment drift can change behavior without code edits. That is not sufficient here: Capabilities and limits vary by model deployment.

B:

Versioned system messages provide a controlled behavioral baseline. That is not sufficient here: Capabilities and limits vary by model deployment.

D:

The inference service needs task content to generate a relevant answer. That is not sufficient here: Capabilities and limits vary by model deployment.

Question 554

A solution team has selected this control: Choose a compatible multimodal model and provide supported text/image inputs. Which observation is the most direct acceptance evidence?

  1. The response includes the requested image artifact and metadata
  2. Repeated structured cases follow the expected schema more consistently
  3. The model answers a question using the actual supplied photo and text
  4. The same brief yields a wider range of acceptable styles without broken constraints

Explanation

Correct Answer

C:

Multimodal models can reason over more than one supported input modality. The expected acceptance evidence is: The model answers a question using the actual supplied photo and text. This observation directly tests the stated behavior.

Incorrect Answers

A:

Text generation and image generation are distinct model output capabilities. This observation does not establish the required result: The model answers a question using the actual supplied photo and text.

B:

Lower randomness often improves consistency but does not guarantee correctness. This observation does not establish the required result: The model answers a question using the actual supplied photo and text.

D:

More varied token sampling can increase creative alternatives. This observation does not establish the required result: The model answers a question using the actual supplied photo and text.

Question 555

The approved design action is: Detect individual package instances and count nonduplicate boxes. Which test result would BEST establish that it worked?

  1. Disallowed outputs are blocked in adversarial evaluation
  2. A labeled photograph receives an accurate high-level summary
  3. The count matches annotated packages after overlap filtering
  4. The crop keeps the subject in view at the specified aspect ratio

Explanation

Correct Answer

C:

Object counts require resolving distinct instances, not trusting one scene label. The expected acceptance evidence is: The count matches annotated packages after overlap filtering. This observation directly tests the stated behavior.

Incorrect Answers

A:

Generative media needs policy controls in addition to prompt instructions. This observation does not establish the required result: The count matches annotated packages after overlap filtering.

B:

Captioning produces a sentence-level description rather than a coordinate list. This observation does not establish the required result: The count matches annotated packages after overlap filtering.

D:

Smart cropping attempts to retain salient regions within target dimensions. This observation does not establish the required result: The count matches annotated packages after overlap filtering.

Question 556

Troubleshooting report: The sample code commits a resource key into a public repository. Which correction is MOST directly targeted to this failure?

  1. Benchmark candidate deployments and choose by requirements
  2. Construct messages in the documented SDK request format
  3. Rotate the key and load credentials through a governed secret mechanism
  4. Prune irrelevant history and preserve the minimum required state

Explanation

Correct Answer

C:

This correction directly addresses the reported failure. Credentials should not appear in repositories, prompt logs or client binaries. The targeted change is: Rotate the key and load credentials through a governed secret mechanism.

Incorrect Answers

A:

Lower-cost deployment is acceptable only if accuracy and safety floors remain met. In the stated case, troubleshooting report: The sample code commits a resource key into a public repository; “Benchmark candidate deployments and choose by requirements” does not target that failure.

B:

Chat APIs use structured messages rather than concatenating untyped command strings. In the stated case, troubleshooting report: The sample code commits a resource key into a public repository; “Construct messages in the documented SDK request format” does not target that failure.

D:

Long irrelevant context consumes tokens and may increase latency. In the stated case, troubleshooting report: The sample code commits a resource key into a public repository; “Prune irrelevant history and preserve the minimum required state” does not target that failure.

Question 557

An acceptance tester needs to validate: Use word or utterance timestamps from a supported recognition output. Which observed result proves the intended behavior?

  1. Subtitle cues match the spoken phrases at measured times
  2. Labeled specialty phrases are transcribed with reduced errors
  3. The selected configuration has lower WER on the same dataset
  4. The selected voice reads the intended language correctly

Explanation

Correct Answer

A:

Timecodes align text segments to audio playback. The expected acceptance evidence is: Subtitle cues match the spoken phrases at measured times. This observation directly tests the stated behavior.

Incorrect Answers

B:

Domain-specific words may require biasing or adaptation beyond a generic recognizer. This observation does not establish the required result: Subtitle cues match the spoken phrases at measured times.

C:

Comparable labeled audio prevents misleading quality claims. This observation does not establish the required result: Subtitle cues match the spoken phrases at measured times.

D:

Not every voice is supported for all languages. This observation does not establish the required result: Subtitle cues match the spoken phrases at measured times.

Question 558

A project needs the following AI behavior: A contract processor needs to retrieve named clauses and page citations. What should the designer implement?

  1. Minimize extracted PII and restrict diagnostic payload storage
  2. Use document extraction with source positions before indexing for search
  3. Choose media analyzers with transcript and scene evidence tied to the retrieved answer
  4. Extract currency code and monetary amount as separate validated fields

Explanation

Correct Answer

B:

Structured layout and provenance let users verify a quoted clause. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Inference outputs and telemetry can contain protected identity information. That is not sufficient here: Structured layout and provenance let users verify a quoted clause.

C:

Multimodal grounding should preserve how the answer maps to source media. That is not sufficient here: Structured layout and provenance let users verify a quoted clause.

D:

A currency symbol or amount alone may not establish financial meaning. That is not sufficient here: Structured layout and provenance let users verify a quoted clause.

Question 559

Troubleshooting report: The app starts every turn with no conversation context. Which correction is MOST directly targeted to this failure?

  1. Update configuration and run supported compatibility checks
  2. Reconcile dependent configuration and rerun targeted regressions
  3. Retain bounded conversation state with relevant history
  4. Attach idempotency keys and verify tool receipts after reconnecting

Explanation

Correct Answer

C:

This correction directly addresses the reported failure. Stateful conversations require supplying or retrieving relevant prior context. The targeted change is: Retain bounded conversation state with relevant history.

Incorrect Answers

A:

A model family label is not necessarily the concrete deployed endpoint. In the stated case, troubleshooting report: The app starts every turn with no conversation context; “Update configuration and run supported compatibility checks” does not target that failure.

B:

Environment drift can change behavior without code edits. In the stated case, troubleshooting report: The app starts every turn with no conversation context; “Reconcile dependent configuration and rerun targeted regressions” does not target that failure.

D:

Client retries should distinguish read requests from side-effecting tool calls. In the stated case, troubleshooting report: The app starts every turn with no conversation context; “Attach idempotency keys and verify tool receipts after reconnecting” does not target that failure.

Question 560

Troubleshooting report: The model invents a required parameter name not in the contract. Which correction is MOST directly targeted to this failure?

  1. Constrain the call to the documented JSON input schema
  2. Partition conversation context by identity and expiration
  3. Exercise negative and expected tool paths before release
  4. Require real tool calls and distinguish pending from completed

Explanation

Correct Answer

A:

This correction directly addresses the reported failure. Invalid tool parameters break calls even if the intended operation is correct. The targeted change is: Constrain the call to the documented JSON input schema.

Incorrect Answers

B:

Conversation continuity should preserve identifiers without crossing user sessions. In the stated case, troubleshooting report: The model invents a required parameter name not in the contract; “Partition conversation context by identity and expiration” does not target that failure.

C:

An answer that sounds plausible may conceal failed tool calls. In the stated case, troubleshooting report: The model invents a required parameter name not in the contract; “Exercise negative and expected tool paths before release” does not target that failure.

D:

Generated text does not prove external execution. In the stated case, troubleshooting report: The model invents a required parameter name not in the contract; “Require real tool calls and distinguish pending from completed” does not target that failure.

Microsoft Azure AI Fundamentals (AI-901) Practice Tests