AI-900 Practice Questions
Prepare for AI-900 with more than an answer.
Unlock the full exam and previous versions
- v1Microsoft Azure AI Fundamentals 150 questions Locked
- AI-900Legacy Azure AI Fundamentals 249 questions Current
- Exam fee
- $99 USD
- Level
- Fundamentals
- Valid for
- Does not expire
Domains covered on the exam 5
- Describe Artificial Intelligence workloads and considerations20%
- Describe fundamental principles of machine learning on Azure17.5%
- Describe features of computer vision workloads on Azure17.5%
- Describe features of Natural Language Processing (NLP) workloads on Azure17.5%
- Describe features of generative AI workloads on Azure22.5%
- 1
A software company is developing a system that automatically generates programming code snippets based on natural language descriptions. For example, a user might type "create a Python function to sort a list of numbers." This is a primary use case for which type of AI workload?
Show answer details
Correct answer: D
Generative AI is a type of artificial intelligence that can create new content, such as text, images, music, and code. Models like OpenAI's Codex, available through Azure OpenAI Service, are specifically trained for code generation tasks based on natural language prompts.
- 2
An organization is building a system that must be auditable, where decisions made by the AI can be traced back and explained to regulators. The team is required to document the model's intended use, performance metrics, and fairness assessments. This practice of documenting a model's characteristics is a key part of which two Responsible AI principles? (Select TWO)
Show answer details
Correct answer: A, C
Transparency is about making the AI system understandable. Comprehensive documentation, such as model cards, directly contributes to transparency by clearly stating what the model does, how it performs, and its limitations.
Accountability involves having mechanisms for responsibility and governance. Thorough documentation is essential for accountability, as it provides the basis for audits, reviews, and assigning responsibility for the system's behavior and impact.
- 3
A retail company uses Azure Machine Learning to train a model that predicts daily sales for each of its stores. The model is deployed to a real-time endpoint. The company needs to monitor the deployed model to detect if the incoming data for prediction starts to differ significantly from the data the model was trained on. This concept is known as:
Show answer details
Correct answer: B
Data drift is the phenomenon where the statistical properties of the data used for real-time prediction change over time compared to the training data. Azure Machine Learning provides tools to monitor for data drift, which can degrade model performance and indicate that the model needs to be retrained.
- 4
True or False: The 'fine-tuning' process for a large language model in Azure OpenAI Service involves retraining the entire model from scratch using a custom dataset.
Show answer details
Correct answer: B
False. Fine-tuning does not retrain the model from scratch. Instead, it takes a pre-trained foundation model (like GPT-3.5-Turbo) and further trains it on a smaller, specific dataset. This process adjusts the model's weights to specialize its performance for a particular task without the massive computational cost of training from zero.
- 5
An educational institution wants to create an interactive learning module that can listen to a student's pronunciation of foreign language vocabulary and provide real-time feedback on accuracy. Which specific capability of the Azure AI Speech service should they use?
Show answer details
Correct answer: D
The Pronunciation Assessment feature of Azure AI Speech is specifically designed for this scenario. It evaluates speech and provides feedback on the accuracy and fluency of pronunciation, making it ideal for language learning applications.
- 6
A financial services firm is developing a generative AI application using Azure OpenAI Service to summarize lengthy financial reports. To ensure the summaries are grounded in the provided document and avoid fabricating information, the development team implements a Retrieval-Augmented Generation (RAG) pattern. Which component of the RAG pattern is responsible for finding the most relevant sections of the source document to guide the language model?
Show answer details
Correct answer: C
In a Retrieval-Augmented Generation (RAG) pattern, the indexing and retrieval system (often a vector database like Azure AI Search) is responsible for searching the source data to find the most relevant information based on the user's query. This retrieved context is then passed to the LLM to generate a grounded response.
- 7
An AI development team is using Azure AI Content Safety to moderate user-generated content for a new social media application. They need to configure the service to detect and filter content related to self-harm and violence with the highest sensitivity. Which two configuration settings are most critical for achieving this requirement? (Select TWO)
Show answer details
Correct answer: A, C
Severity levels (e.g., 0-7) allow for fine-grained control over how sensitive the detection is for each category. Setting a low threshold for the 'self-harm' and 'violence' categories makes the filtering more aggressive.
Custom blocklists allow the team to add specific terms, phrases, or patterns that are highly indicative of the content they wish to filter, providing an extra layer of targeted moderation.
- 8
True or False: The Azure AI Foundry model catalog exclusively contains proprietary models developed by Microsoft.
Show answer details
Correct answer: B
The Azure AI Foundry model catalog is designed to be comprehensive and includes a wide variety of models, including state-of-the-art open-source models (like those from Hugging Face and Meta), models from Microsoft, and models from OpenAI.
- 9
A hospital is implementing an AI system to analyze medical text from patient records. The system needs to identify mentions of specific medical conditions, medications, and dosages. This information will be used to populate a structured database for clinical research. Which feature of the Azure AI Language service is specifically designed for this task?
Show answer details
Correct answer: C
Custom Named Entity Recognition (NER) is the ideal feature for this scenario. While standard NER can find general entities like people and places, Custom NER allows the hospital to train a model to recognize specific, domain-relevant entities such as 'medical conditions', 'medications', and 'dosages'.
- 10
A global news organization wants to create a service that provides near real-time translation of breaking news articles into multiple languages for its international audience. The solution must preserve the original document's formatting and structure. Which Azure AI service is the most appropriate choice for this requirement?
Show answer details
Correct answer: D
Azure AI Translator, specifically its Document Translation feature, is designed for this exact purpose. It can translate entire documents or batches of documents in various file formats while preserving the original structure and formatting, which is a key requirement for the news organization.
