Free AI-901 practice test: 30 questions with answers

Large language models generate and transform text in response to prompts, which covers drafting, answering, and summarizing. Packet routing, face detection, and invoice printing are not what these models do.

Azure OpenAI Service gives access to OpenAI models with Azure security and governance features. Cosmos DB, Azure Firewall, and Azure Backup provide storage, network protection, and backup rather than generative models.

A prompt is the input of instructions and context that guides a generative model's response. Protocols, dashboards, and storage containers are other parts of an application, not the prompt.

Azure AI Foundry is the platform that brings together a model catalogue, prompt tools, and development resources for generative AI. Monitor, Event Grid, and DNS support monitoring, eventing, and naming rather than model development.

A copilot is an AI assistant embedded in an application to help the user with tasks such as drafting, summarizing, or answering questions. Aircraft, VPNs, and database indexes are unrelated to this term.

Content filtering and human review catch harmful or inaccurate output before it reaches users. Publishing every prompt, disabling logs, and skipping output checks each weaken safety or accountability.

Key phrase extraction returns the main concepts of a text, which is how talking points are identified. Speech synthesis, image classification, and forecasting work on different types of input or produce different outputs.

Sentiment analysis scores the opinion expressed in text, which is what the question describes. Entity recognition, object detection, and key phrase extraction answer different questions about the same text or about images.

Entity recognition detects and categorizes named items in text, which is what the question asks for. Speech recognition, sentiment analysis, and anomaly detection do not identify named entities.

Azure AI Speech provides speech to text and text to speech capabilities. Azure AI Vision analyses images, while Azure Monitor and Cosmos DB handle telemetry and data storage.

Speech synthesis, also called text to speech, produces spoken audio from text. Speech recognition goes the other way, while key phrase extraction and image classification do not generate audio.

Question answering and custom named entity recognition are text capabilities offered by Azure AI Language. Image tagging belongs to computer vision, while forecasting and anomaly detection on numeric series belong to Azure machine learning and analytics tools.

Regression predicts continuous values, which is what a house price is. Clustering groups without labels, classification predicts categories, and reinforcement learning learns from rewards rather than predicting a value directly.

A label is the known outcome for each training example, and supervised learning uses it to learn the mapping from features. File names, identifier columns, and chart titles are not labels.

Automated machine learning in Azure Machine Learning searches across algorithms and settings and ranks the resulting models. Cosmos DB, Logic Apps, and Azure Monitor provide data storage, workflow automation, and monitoring rather than model search.

A validation dataset is held apart from training to tune hyperparameters and detect overfitting before the final test. Backups, labelling test data, and charting are not its purpose.

Clustering and dimensionality reduction work on unlabelled data to find structure. Regression, classification, and labelled forecasting all require known target values, which places them in supervised learning.

Features are the input variables from which the model learns to predict the label. Collector names, charts, and billing subscriptions are not features of the data.

Image classification labels the whole image, while object detection locates objects, OCR extracts text, and face detection finds faces. Only image classification matches the question as asked.

Object detection returns both the objects and their bounding box locations, which is what the question describes. Classification gives no positions, colour analysis reports colours, and captioning gives text without coordinates.

Azure AI Face provides face detection and analysis, subject to responsible use policies. Azure AI Language, Cosmos DB, and Event Hubs handle text, data storage, and event streams rather than faces in images.

Optical character recognition extracts text from images, which makes the content searchable and storable. Sentiment analysis, translation, and anomaly detection process text or numbers that have already been extracted.

Fairness across groups and consent and privacy are central responsible AI concerns for facial analysis. Shutter speed, fonts, and box colours affect usability or image quality, not the ethical and fairness issues the solution raises.

Image tagging returns descriptive words for what an image contains, such as objects, scenes, and attributes. Anomaly detection, language modelling, and speech recognition do not analyse image content.

Optical character recognition is a computer vision workload that converts printed text in images into machine-readable text. Natural language processing works on text that already exists, while anomaly detection and forecasting work on numeric series.

Fairness is the principle concerned with equitable treatment and avoiding unjustified differences in outcomes across groups. Transparency, reliability and safety, and accountability are important principles, but each addresses a different concern.

Anomaly detection finds values that depart from an expected pattern, which is what unusual transactions look like against a customer's history. Object detection, translation, and speech synthesis solve different problems.

Transparency is the principle that people can understand how the system works and reaches its decisions. Accountability, privacy and security, and inclusiveness are separate principles that address other concerns.

Accountability keeps people responsible for AI systems, and privacy and security protect data and the people it describes. Removing demographic data does not guarantee fairness, reliability applies to software as well as robots, and transparency generally requires documentation.

Document processing extracts fields, tables, and key-value pairs from forms such as invoices, which is the structured extraction the question describes. Image classification, speech synthesis, and sentiment analysis solve different problems.