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 61

Question 600

An acceptance tester needs to validate: Use retrieval-augmented generation with actual source citations. Which observed result proves the intended behavior?

  1. The spoken message matches the approved written text
  2. The minutes correctly associate tasks with speakers and source timestamps
  3. The result contains product boxes in their correct shelf positions
  4. Answers cite valid manual passages and avoid unsupported details

Explanation

Correct Answer

D:

RAG retrieves current external evidence for a generated response. The expected acceptance evidence is: Answers cite valid manual passages and avoid unsupported details. This observation directly tests the stated behavior.

Incorrect Answers

A:

Speech synthesis converts supplied text into audible output. This observation does not establish the required result: Answers cite valid manual passages and avoid unsupported details.

B:

Attribution distinguishes who committed to each action. This observation does not establish the required result: Answers cite valid manual passages and avoid unsupported details.

C:

Object detectors return classes and spatial boundaries. This observation does not establish the required result: Answers cite valid manual passages and avoid unsupported details.

Question 601

An Azure AI solution uses the following approved control: Use sentiment analysis to estimate positive, negative, or neutral tone. What observable result should pass QA?

  1. A logged tool receipt proves a permitted ticket was created
  2. Held-out requests receive the appropriate routed intent labels
  3. The narrated explanation matches labeled chart features without inventing values
  4. A labeled review set produces measured sentiment predictions

Explanation

Correct Answer

D:

Sentiment analysis infers expressed attitude rather than extracting names. The expected acceptance evidence is: A labeled review set produces measured sentiment predictions. This observation directly tests the stated behavior.

Incorrect Answers

A:

Agentic workloads combine reasoning and bounded external actions. This observation does not establish the required result: A labeled review set produces measured sentiment predictions.

B:

Classification maps text to one of several predefined categories. This observation does not establish the required result: A labeled review set produces measured sentiment predictions.

C:

Multimodal input can describe the graphic and speech output can read that description. This observation does not establish the required result: A labeled review set produces measured sentiment predictions.

Question 602

A project needs the following AI behavior: A retailer has unlabeled shopping transactions and wants groups of similar customers. What should the designer implement?

  1. Use retrieval-augmented generation from current approved catalog sources
  2. Confirm the model deployment identifier and endpoint resource match
  3. Use clustering or another appropriate unsupervised technique
  4. Choose a speech-recognition model or audio-capable multimodal endpoint with a documented audio-input contract

Explanation

Correct Answer

C:

Unsupervised learning finds structure without supplied target labels. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

External retrieval updates factual context without retraining model weights. That is not sufficient here: Unsupervised learning finds structure without supplied target labels.

B:

Applications call a deployed model resource rather than a generic marketing model name. That is not sufficient here: Unsupervised learning finds structure without supplied target labels.

D:

Model capabilities must include the required audio modality before the client can obtain speech-to-text results. That is not sufficient here: Unsupervised learning finds structure without supplied target labels.

Question 603

A project needs the following AI behavior: A product team needs to decide whether a generative model is suitable for deterministic tax arithmetic. What should the designer implement?

  1. Select an image-generation-capable model and a supported image endpoint
  2. Choose a text-classification or sentiment-analysis capability rather than an image generator
  3. Use retrieval-augmented generation from current approved catalog sources
  4. Use a conventional calculation function for exact arithmetic and let the model explain it

Explanation

Correct Answer

D:

Generative models are not a substitute for auditable deterministic computation. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Text generation and image generation are distinct model output capabilities. That is not sufficient here: Generative models are not a substitute for auditable deterministic computation.

B:

Different model tasks require compatible outputs and labeled evaluation data. That is not sufficient here: Generative models are not a substitute for auditable deterministic computation.

C:

External retrieval updates factual context without retraining model weights. That is not sufficient here: Generative models are not a substitute for auditable deterministic computation.

Question 604

The approved design action is: Check feature, model and regional availability before creating an analyzer. Which test result would BEST establish that it worked?

  1. The result is read only after the operation succeeds
  2. Reviewers can inspect the handwritten comment's location
  3. The chosen analyzer runs in a supported approved region
  4. The lot code matches the visible label and its image region

Explanation

Correct Answer

C:

Some modalities and features depend on resource and API version. The expected acceptance evidence is: The chosen analyzer runs in a supported approved region. This observation directly tests the stated behavior.

Incorrect Answers

A:

Long-running media analysis may be asynchronous. This observation does not establish the required result: The chosen analyzer runs in a supported approved region.

B:

Source-location evidence helps verify recognition on mixed-layout documents. This observation does not establish the required result: The chosen analyzer runs in a supported approved region.

D:

Visual information extraction can map pixels to structured identifiers. This observation does not establish the required result: The chosen analyzer runs in a supported approved region.

Question 605

A project needs the following AI behavior: A support team wants the most important phrases from long feedback messages. What should the designer implement?

  1. Use object detection when each product needs a location
  2. Use text embeddings with nearest-neighbor search and business filters
  3. Use key phrase extraction to identify salient text spans
  4. Use text summarization with fidelity checks against sources

Explanation

Correct Answer

C:

Key phrase extraction finds prominent topics without inventing new source material. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Object detectors return classes and spatial boundaries. That is not sufficient here: Key phrase extraction finds prominent topics without inventing new source material.

B:

Semantic vectors capture related concepts while filters enforce eligibility. That is not sufficient here: Key phrase extraction finds prominent topics without inventing new source material.

D:

Summarization compresses content without independently validating missing facts. That is not sufficient here: Key phrase extraction finds prominent topics without inventing new source material.

Question 606

An AI operations team is testing two distinct capabilities. Failure (1): The app parses free-form prose and assumes all required keys exist. Failure (2): The app scores fluency as proof of correctness. Which TWO specific corrections should the team apply?

  1. Increase the supported limit within deployment quotas or shorten the requested output
  2. Verify claims against trusted evidence and permit abstention
  3. Reduce sampling variability and validate outputs against a schema
  4. Define allowed JSON keys and validate the returned types
  5. Apply context selection and retain necessary evidence only

Explanation

Correct Answers

B:

Correct for failure (2). Language-model likelihood does not independently establish factual truth. It independently corrects the second defect.

D:

Correct for failure (1). Natural-language fluency is not a guarantee of machine-readable contract validity. The proposed correction directly addresses that reported defect.

Incorrect Answers

A:

Generation stops when an output budget or stop condition is reached. The stated pair of failures instead requires “Verify claims against trusted evidence and permit abstention” and “Define allowed JSON keys and validate the returned types”; this option cannot substitute for both.

C:

Lower randomness often improves consistency but does not guarantee correctness. The stated pair of failures instead requires “Verify claims against trusted evidence and permit abstention” and “Define allowed JSON keys and validate the returned types”; this option cannot substitute for both.

E:

Processing extra tokens increases cost and can affect response time. The stated pair of failures instead requires “Verify claims against trusted evidence and permit abstention” and “Define allowed JSON keys and validate the returned types”; this option cannot substitute for both.

Question 607

A project needs the following AI behavior: A business wants a searchable knowledge base from long technical manuals. What should the designer implement?

  1. Use layout-aware document analysis that preserves rows, selection marks and text spans
  2. Use an appropriate search-oriented document analyzer with meaningful chunk boundaries
  3. Define a custom analyzer for consistent target field meanings across templates
  4. Choose media analyzers with transcript and scene evidence tied to the retrieved answer

Explanation

Correct Answer

B:

RAG preparation should preserve document structure and source references. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Ordinary plain-text OCR may lose table relationships and checkbox states. That is not sufficient here: RAG preparation should preserve document structure and source references.

C:

A schema-driven analyzer identifies business fields independent of fixed coordinates. That is not sufficient here: RAG preparation should preserve document structure and source references.

D:

Multimodal grounding should preserve how the answer maps to source media. That is not sufficient here: RAG preparation should preserve document structure and source references.

Question 608

Troubleshooting report: The search index exposes all extracted pages to every client. Which correction is MOST directly targeted to this failure?

  1. Validate field presence against source evidence
  2. Use table-aware layout extraction and numeric validation
  3. Run a supported OCR or visual text extraction feature
  4. Enforce source permissions in indexing and query-time filtering

Explanation

Correct Answer

D:

This correction directly addresses the reported failure. Extracting content does not waive document access controls. The targeted change is: Enforce source permissions in indexing and query-time filtering.

Incorrect Answers

A:

Extraction should preserve absence and distinguish it from zero. In the stated case, troubleshooting report: The search index exposes all extracted pages to every client; “Validate field presence against source evidence” does not target that failure.

B:

Cell relationships are necessary to check item totals. In the stated case, troubleshooting report: The search index exposes all extracted pages to every client; “Use table-aware layout extraction and numeric validation” does not target that failure.

C:

OCR is appropriate for characters embedded in a photograph. In the stated case, troubleshooting report: The search index exposes all extracted pages to every client; “Run a supported OCR or visual text extraction feature” does not target that failure.

Question 609

An Azure AI Fundamentals solution has this requirement: An accessibility app needs detailed descriptions of several image regions. Which implementation most directly addresses it?

  1. Provide both images to an appropriate multimodal comparison workflow
  2. Use a supported dense-captioning capability where available
  3. Request an image caption or supported multimodal description
  4. Use region-aware captioning or a cropped image focused on that location

Explanation

Correct Answer

B:

Dense captions describe localized regions beyond a single global caption. This implementation addresses the precise goal rather than an adjacent AI task.

Incorrect Answers

A:

Change detection needs both temporal states for a meaningful comparison. That is not sufficient here: Dense captions describe localized regions beyond a single global caption.

C:

Captioning produces a sentence-level description rather than a coordinate list. That is not sufficient here: Dense captions describe localized regions beyond a single global caption.

D:

Specific regions can carry details missed by a whole-image caption. That is not sufficient here: Dense captions describe localized regions beyond a single global caption.

Question 610

Troubleshooting report: Overlapping detections are counted twice. Which correction is MOST directly targeted to this failure?

  1. Gate distribution on the image moderation result
  2. Add confidence gating and supporting evidence
  3. Apply confidence and nonmaximum suppression or instance deduplication
  4. Route to the text-to-image mode with intended inputs only

Explanation

Correct Answer

C:

This correction directly addresses the reported failure. Object counts require resolving distinct instances, not trusting one scene label. The targeted change is: Apply confidence and nonmaximum suppression or instance deduplication.

Incorrect Answers

A:

Generative media needs policy controls in addition to prompt instructions. In the stated case, troubleshooting report: Overlapping detections are counted twice; “Gate distribution on the image moderation result” does not target that failure.

B:

Generated visual interpretations can be wrong even when phrased fluently. In the stated case, troubleshooting report: Overlapping detections are counted twice; “Add confidence gating and supporting evidence” does not target that failure.

D:

Generation from text is distinct from image editing. In the stated case, troubleshooting report: Overlapping detections are counted twice; “Route to the text-to-image mode with intended inputs only” does not target that failure.

Microsoft Azure AI Fundamentals (AI-901) Practice Tests