Free AI-900 practice test

Answer: D. Fairness means similar people should be treated similarly, and here outcomes differ by group for no valid reason. Transparency is about explaining how the system works, inclusiveness about serving people of all abilities, and reliability about consistent, safe operation.

Answer: C. Telling users they are dealing with an AI and what it can and cannot do is transparency. Accountability is about people answering for the system, privacy about protecting data, and fairness about equal treatment.

Answer: B. Inclusiveness means designing AI so it benefits and can be used by people with a wide range of abilities. The other principles cover safe operation, human responsibility and explainability, not accessibility.

Answer: B. Generative models predict plausible next tokens, so without grounding in real company data they can produce fluent but false answers. Grounding the model in the actual refund policy reduces this. A temperature of zero makes output more consistent but does not make it factual.

Answer: B. A lower temperature makes the model choose more likely tokens, so answers become more predictable. A higher temperature adds variety, and removing the system message or adding users does nothing for consistency.

Answer: A. The app must understand an image and a written question together, which is what a multimodal model does. Translation and speech synthesis models do not interpret images, and a price regression model solves an unrelated problem.

Answer: D. Entity detection finds and labels items such as organizations, dates and places in text. Sentiment analysis judges opinion, speech synthesis produces audio, and image classification works on pictures.

Answer: C. An agent combines a model with tools so it can retrieve data and perform actions across several steps. OCR reads text from images, speech recognition only transcribes audio, and image generation creates pictures.

Answer: B. The system message sets the model's behaviour, tone and boundaries for the whole conversation. Deployment names, profile pictures and subscription names have no effect on how the model responds.

Answer: A. Apps call a deployed model through its deployment, so a model that has only been viewed in the catalog cannot be used yet. Restarting, changing language or adjusting temperature do not create a deployment.

Answer: A. Chat model calls are stateless, so the client must include the conversation history in each request for the model to keep context. Font size and temperature do not add memory, and new deployments would not carry history either.

Answer: D. Giving the agent a retrieval tool over the policy documents grounds its answers in the real content. More creativity increases the risk of invented answers, image generation is unrelated, and a word limit does not make answers accurate.

Answer: C. The Foundry portal lets you test an agent interactively before building an app around it. A spreadsheet or printout cannot run the agent, and the billing page only shows costs.

Answer: B. Sentiment analysis classifies the opinion expressed in text, which is exactly the requirement. Speech synthesis creates audio, image generation creates pictures, and OCR reads text from images rather than judging opinion.

Answer: B. Turning written text into spoken audio is speech synthesis. Speech recognition does the reverse, and entity detection and image classification do not produce audio.

Answer: B. A multimodal model that accepts audio can take the spoken prompt directly and respond. A text-only model needs the speech transcribed first, and image generation or sentiment analysis cannot interpret spoken questions.

Answer: A. Multimodal chat models accept image content alongside text in the same message, so the model can reason about both. Deployment names carry no content, and audio or speech services cannot interpret the photo.

Answer: A. Creating new pictures from text prompts is the job of an image generation model. The other models handle speech, translation or text analysis and cannot produce images.

Answer: A. Content Understanding extracts information from documents and forms into structured output based on the fields you define. A chat model without the files cannot see them, and speech synthesis or image generation do not extract data.

Answer: C. Content Understanding can process audio and video and return structured information such as summaries and defined fields. Speech synthesis produces audio rather than analysing it, image classification does not work on audio, and storing recordings alone meets none of the needs.