AI-900 Exam Dumps - Microsoft Certified: Azure AI Fundamentals Questions and Answers

The correct answer is A. Configure the Content filters settings.

When using the Azure OpenAI Service for text or image generation, Microsoft provides built-in content filtering to help detect and block potentially harmful or unsafe outputs. These filters are part of Microsoft’s Responsible AI framework and are designed to prevent the generation of offensive, violent, sexual, or otherwise restricted content.

To ensure the highest level of protection, you can configure content filter settings within the Azure OpenAI deployment. This allows you to define stricter policies based on your organization’s safety requirements. For image generation models such as DALL·E, enabling or strengthening these filters ensures that inappropriate or unsafe images are not generated or returned.

B (Customize an LLM): Customization affects behavior but not safety filtering.

C (Configure the system prompt): Adjusts response style but doesn’t guarantee content safety.

D (Change the model): Different models have similar filter systems; protection level depends on filter configuration, not the model itself.

Box 1: Select Columns in Dataset

For Columns to be cleaned, choose the columns that contain the missing values you want to change. You can choose multiple columns, but you must use the same replacement method in all selected columns.

Example:

The task is to build a machine learning model in Azure Machine Learning designer to predict automobile prices, which is a regression problem since the output (price) is a continuous numeric value. The pipeline must follow the logical data preparation, training, and evaluation flow as outlined in the Microsoft Azure AI Fundamentals (AI-900) study guide and Microsoft Learn module “Create a machine learning model with Azure Machine Learning designer.”

Here’s the correct sequence and reasoning:

Select Columns in Dataset:The first step after loading the raw automobile dataset is to choose the relevant columns that will be used as features (inputs) and the label (output). This module ensures that only necessary fields (for example, horsepower, engine size, mileage, etc.) are used to train the model while excluding irrelevant columns like vehicle ID or serial number.

Split Data:Next, the cleaned and filtered dataset must be split into two subsets: training data and testing data (often 70/30 or 80/20). This allows the model to be trained on one portion and evaluated on the other to measure predictive accuracy.

Linear Regression:Since automobile price prediction is a numeric prediction task, the appropriate learning algorithm is Linear Regression. This supervised algorithm learns relationships between numeric features and the target (price).

Finally, the workflow connects the training data and Linear Regression module to the Train Model module, which outputs a trained regression model. The trained model is then linked to the Score Model module to compare predicted vs. actual prices.

This pipeline fully aligns with Microsoft’s recommended process for regression in Azure ML Designer.

When using generative AI models such as those in Azure OpenAI, it’s critical to monitor and control the model’s outputs to prevent the generation of harmful, hateful, or unsafe content. According to Microsoft’s Responsible AI guidelines and Azure OpenAI safety systems, this is achieved through content filtering mechanisms.

Content filtering automatically analyzes generated text and blocks or flags responses containing hate speech, harassment, sexual content, or violent language. Azure OpenAI applies these filters by default through its built-in content moderation system, which categorizes responses into safety levels (safe, review, block).

Option review:

A. Prompt engineering: Helps guide responses but does not detect hate content automatically.

B. Abuse monitoring: Refers to manual or post-processing review, not real-time automated filtering.

C. Content filtering: ✅ Correct — actively detects and prevents hateful or unsafe responses.

D. Fine-tuning: Adjusts model behavior using additional training data but doesn’t ensure safety monitoring.

Box 1: Entity recognition

the Named Entity Recognition module in Machine Learning Studio (classic), to identify the names of things, such as people, companies, or locations in a column of text.

Named entity recognition is an important area of research in machine learning and natural language processing (NLP), because it can be used to answer many real-world questions, such as:

Which companies were mentioned in a news article?

Does a tweet contain the name of a person? Does the tweet also provide his current location?

Were specified products mentioned in complaints or reviews?

Box 2: Sentiment Analysis

The Text Analytics API ' s Sentiment Analysis feature provides two ways for detecting positive and negative sentiment. If you send a Sentiment Analysis request, the API will return sentiment labels (such as " negative " , " neutral " and " positive " ) and confidence scores at the sentence and document-level.

In a chat solution using the Azure OpenAI GPT-3.5 model, the temperature parameter controls the creativity and variability of generated responses. According to the Microsoft Learn documentation for Azure OpenAI Service, temperature is a float value typically between 0 and 2, determining how deterministic or random the model’s output is. A lower temperature (e.g., 0–0.3) makes responses more focused and deterministic, while a higher temperature (e.g., 0.8–1.2) produces more verbose, creative, and diverse responses.

When you want the chat model to generate more detailed or expressive output, increasing the temperature encourages the model to explore a broader range of possible continuations, leading to longer and more varied text. This parameter directly affects how “verbose” or elaborate the model’s responses can be, which is why it is the correct answer.

The other options are not appropriate for this scenario:

A. Presence penalty reduces repetition by discouraging reuse of the same phrases but does not control verbosity.

C. Stop sequence defines tokens where generation should stop, limiting rather than extending response length.

D. Max response (max tokens) controls the maximum length of the response but does not inherently make answers more verbose or expressive.

Thus, to encourage more elaborate and detailed output from the Azure OpenAI GPT-3.5 model, the correct configuration parameter to adjust is Temperature (B).

The correct answer is object detection. According to the Microsoft Azure AI Fundamentals (AI-900) official study materials and Microsoft Learn module “Explore computer vision”, object detection is the process of identifying and locating objects within an image or video. The primary characteristic of object detection, as emphasized in the study guide, is its ability to return a bounding box around each detected object along with a corresponding label or class.

In this question, the task involves returning a bounding box that indicates the location of a vehicle in an image. This is the exact definition of object detection — identifying that the object exists (a vehicle) and determining its position within the frame. Microsoft Learn clearly differentiates this from other computer vision tasks. Image classification, for example, only determines what an image contains as a whole (for instance, “this image contains a vehicle”), but it does not indicate where in the image the object is located. Optical character recognition (OCR) is specifically used for extracting printed or handwritten text from images, and semantic segmentation involves classifying every pixel in an image to understand boundaries in greater detail, often used in autonomous driving or medical imaging.

The official AI-900 guide highlights object detection as one of the key computer vision workloads supported by Azure Computer Vision, Custom Vision, and Azure Cognitive Services. These services are designed to detect multiple instances of various object types in a single image, outputting bounding boxes and confidence scores for each.

Therefore, based on the AI-900 official curriculum and Microsoft Learn concepts, returning a bounding box that shows the location of a vehicle is a textbook example of object detection, as it involves both recognition and localization of the object within the image frame.

Box 1: 11

TP = True Positive.

The class labels in the training set can take on only two possible values, which we usually refer to as positive or negative. The positive and negative instances that a classifier predicts correctly are called true positives (TP) and true negatives (TN), respectively. Similarly, the incorrectly classified instances are called false positives (FP) and false negatives (FN).

Box 2: 1,033

FN = False Negative

Reliability & Safety

https://en.wikipedia.org/wiki/Tay_(bot)

“To build trust, it ' s critical that AI systems operate reliably, safely, and consistently under normal circumstances and in unexpected conditions. These systems should be able to operate as they were originally designed, respond safely to unanticipated conditions, and resist harmful manipulation. It ' s also important to be able to verify that these systems are behaving as intended under actual operating conditions. How they behave and the variety of conditions they can handle reliably and safely largely reflects the range of situations and circumstances that developers anticipate during design and testing. We believe that rigorous testing is essential during system development and deployment to ensure AI systems can respond safely in unanticipated situations and edge cases, don ' t have unexpected performance failures, and don ' t evolve in ways that are inconsistent with original expectations”

According to the Microsoft Azure AI Fundamentals (AI-900) official study guide and the Microsoft Learn module “Identify guiding principles for responsible AI”, responsible AI is built upon six foundational principles: Fairness, Reliability and Safety, Privacy and Security, Inclusiveness, Transparency, and Accountability. Each principle serves to guide the ethical design, deployment, and management of artificial intelligence systems.

Fairness – This principle ensures that AI systems treat all people fairly and do not discriminate based on personal attributes such as gender, race, or age. The Microsoft Learn content emphasizes that “AI systems should treat everyone fairly” and that organizations must evaluate datasets and model outputs for bias. In this scenario, “The system must not discriminate based on gender, race” clearly aligns with Fairness because it directly addresses equitable treatment and unbiased decision-making.

Privacy and Security – Microsoft’s responsible AI framework stresses that “AI systems must be secure and respect privacy.” This means personal data should be safeguarded, processed lawfully, and visible only to authorized users. The statement “Personal data must be visible only to approved users” reflects the importance of protecting sensitive information and controlling access—precisely the intent of the Privacy and Security principle.

Transparency – Transparency refers to ensuring that users understand how AI systems operate and make decisions. Microsoft notes that “AI systems should be understandable and users should be able to know why decisions are made.” The requirement “Automated decision-making processes must be recorded so that approved users can identify why a decision was made” directly supports this principle. Transparency promotes trust and accountability by documenting the reasoning behind AI outputs.

Reliability and Safety, though another core principle, does not directly relate to any of the provided statements in this question.

Full Detailed Explanation (250–300 words):

“You can fine-tune some Azure OpenAI models by using your own data.” – YESThis statement is true. Azure OpenAI allows customers to fine-tune certain models like GPT-3, GPT-3.5, and some embedding models with their own data. Fine-tuning customizes a model to perform better on specific tasks or match a company’s domain terminology, tone, or context. According to Microsoft Learn’s AI-900 and Azure OpenAI documentation, fine-tuning is supported for approved use cases while maintaining Microsoft’s Responsible AI oversight and compliance process.

“Pretrained generative AI models are a component of Azure OpenAI.” – YESThis statement is also true. Azure OpenAI provides access to pretrained large language and generative AI models such as GPT-3.5, GPT-4, Codex, and DALL·E. These models are pretrained on vast datasets and made available via APIs, allowing developers to generate text, code, and images without needing to train their own models. This is a core feature of Azure OpenAI’s service offering.

“To build a solution that complies with Microsoft responsible AI principles, you must build and train your own model.” – NOThis statement is false. Compliance with Microsoft Responsible AI principles (Fairness, Reliability & Safety, Privacy & Security, Inclusiveness, Transparency, Accountability) does not require building custom models. Prebuilt Azure AI and OpenAI services already align with Responsible AI standards. Developers simply need to use these services responsibly, applying governance and ethical design practices.