AI-900 Practice Test 16 — 15 Questions

Question 1

What is 'token budget' and 'context window' management in large language models?

Accepted Answer

The maximum text the model can process in one call, requiring strategic management of what fits in context. 'token budget' and 'context window' refer to the maximum number of tokens (words, subwords, or characters) a large language model can process in a single inference call. The context window is a fixed limit (e.g., 4096 tokens for GPT-3.5, 8192 for GPT-4), and managing the token budget involves strategically truncating, summarizing, or prioritizing input text to fit within this window, ensuring the model can generate coherent and relevant responses without exceeding its capacity.

Answer

The monthly spending limit set for an Azure OpenAI subscription

Answer

The number of API calls allowed per minute before rate limiting kicks in

Answer

A pre-purchase of tokens at a discounted rate for high-volume Azure OpenAI users

Question 2

What is the difference between a chatbot and a conversational AI agent?

Accepted Answer

Chatbots use fixed rules/decision trees; conversational AI uses NLP/ML for flexible, context-aware responses. Chatbots traditionally rely on predefined rules or decision trees to handle user inputs, limiting them to scripted interactions. In contrast, conversational AI agents leverage natural language processing (NLP) and machine learning (ML) to understand intent, manage context, and generate dynamic, human-like responses. This allows conversational AI to handle ambiguous phrasing, maintain multi-turn dialogue state, and adapt to user behavior without explicit programming for every scenario.

Answer

Chatbots are always voice-based; conversational AI is text-only

Answer

Chatbots are more expensive to build than conversational AI

Answer

They are the same technology with different marketing terms

Question 3

A data scientist has a dataset containing thousands of labeled images of cats and dogs. The data scientist wants to train a model that can automatically classify new unlabeled images as either 'cat' or 'dog'. Which type of machine learning should the data scientist use?

Accepted Answer

Supervised learning. Supervised learning, because the dataset contains labeled images (each image is tagged as 'cat' or 'dog'), and the goal is to train a model to predict the label for new unlabeled images. Supervised learning algorithms, such as convolutional neural networks (CNNs), learn a mapping from input features (pixel values) to output labels using the provided ground-truth labels, enabling accurate classification on unseen data.

Answer

Unsupervised learning

Answer

Reinforcement learning

Answer

Semi-supervised learning

Question 4

A business analyst wants to quickly summarize the main topics discussed in a large collection of customer feedback emails. The analyst needs to identify recurring concepts such as 'product quality', 'shipping delay', and 'customer service'. They want to use a prebuilt Azure AI Language feature without any custom training. Which feature should they use?

Accepted Answer

Key Phrase Extraction. Key Phrase Extraction is the correct choice because it is a prebuilt Azure AI Language feature designed to automatically identify and return the main topics, concepts, and recurring themes from unstructured text, such as 'product quality' or 'shipping delay'. Unlike custom-trained models, this feature requires no training data and works out-of-the-box, making it ideal for quickly summarizing large collections of customer feedback emails.

Answer

Named Entity Recognition (NER)

Answer

Language Detection

Answer

Sentiment Analysis

Question 5

A real estate agency wants to create a feature on their website that automatically crops uploaded property photos to focus on the house itself, removing excess sky, ground, or other surroundings. Which Azure Computer Vision capability should they use?

Accepted Answer

Smart cropping. Smart cropping is the correct capability because it uses AI to identify the most visually salient region of an image and automatically crops it to focus on the main subject—in this case, the house—while removing irrelevant background like sky or ground. This is distinct from generic cropping as it leverages computer vision to detect the primary object and compositionally frame it.

Answer

OCR (Optical Character Recognition)

Answer

Image captioning

Answer

Object detection

Question 6

A law firm needs to process thousands of legal contracts. They want to extract specific clauses (e.g., 'confidentiality', 'indemnity') and also identify the names of parties and monetary amounts mentioned. Which Azure AI Language feature(s) should they use together to achieve both tasks?

Accepted Answer

Custom text classification and named entity recognition. Custom text classification allows the law firm to train a model to identify specific clauses like 'confidentiality' and 'indemnity' by providing labeled examples. Named entity recognition (NER) can then extract predefined entities such as person names (parties) and monetary amounts from the text. Together, these two features address both custom clause detection and standard entity extraction.

Answer

Key phrase extraction and sentiment analysis

Answer

Entity linking and language detection

Answer

Summarization and conversational language understanding

Question 7

What is the difference between a binary classification model and a multi-class classification model?

Accepted Answer

Binary classification predicts two outcomes; multi-class predicts three or more outcomes. Binary classification models are designed to predict exactly two possible outcomes (e.g., spam/not spam), while multi-class classification models predict three or more mutually exclusive classes (e.g., classifying images of cats, dogs, and birds). In Azure Machine Learning, binary classification algorithms like Logistic Regression output a single probability score, whereas multi-class algorithms like Multinomial Logistic Regression or One-vs-Rest meta-estimators output a probability distribution across all classes.

Answer

Binary classification uses numeric outputs; multi-class uses categorical outputs

Answer

Binary is for images; multi-class is for text

Answer

Binary classification is always more accurate than multi-class

Question 8

A developer wants to use Azure OpenAI to build a customer service chatbot that can answer questions about a company's return policy. They create a set of example question-answer pairs in the prompt without retraining the model. Which technique is being used?

Accepted Answer

Few-shot learning. Few-shot learning is the correct technique because the developer provides a small set of example question-answer pairs directly in the prompt to guide the model's responses, without retraining or updating the model's weights. This leverages the model's pre-existing knowledge to generalize from the examples, which is a hallmark of few-shot prompting in Azure OpenAI.

Answer

Fine-tuning

Answer

Reinforcement learning

Answer

Transfer learning

Question 9

A news agency receives thousands of articles daily from wire services. They want to automatically identify and tag mentions of people, organizations, and locations within each article to improve search and categorization. The agency has no labeled training data. Which built-in Azure AI Language feature should they use?

Accepted Answer

Named Entity Recognition (NER). Named Entity Recognition (NER) is the correct choice because it is specifically designed to identify and categorize mentions of people, organizations, locations, and other entity types in unstructured text. Since the agency has no labeled training data, NER's pre-trained model can be used out-of-the-box without any custom training, making it ideal for automatically tagging articles with these entity types to improve search and categorization.

Answer

Sentiment Analysis

Answer

Key Phrase Extraction

Answer

Language Detection

Question 10

A healthcare organization uses an AI system to predict patient readmission risk. The model was trained on data from a single hospital with a predominantly elderly population. When deployed to a different hospital with a younger demographic, the model's accuracy drops significantly. Which responsible AI principle is most directly violated?

Accepted Answer

Reliability and safety. The model's accuracy drop when applied to a different demographic indicates a failure in reliability and safety. The model was trained on a non-representative dataset (elderly patients) and does not generalize to younger populations, violating the principle that AI systems must perform consistently and safely across intended deployment contexts.

Answer

Transparency

Answer

Fairness

Answer

Privacy and security

Question 11

What is 'Azure AI Speech Studio' and what does it help you do?

Accepted Answer

A no-code web portal for testing and configuring Azure AI Speech capabilities. Azure AI Speech Studio is a no-code web portal that allows you to test, configure, and integrate Azure AI Speech capabilities—such as speech-to-text, text-to-speech, custom voice, and speech translation—without writing code. It provides a graphical interface to experiment with prebuilt models, tune recognition accuracy, and generate sample code for deployment, making it ideal for rapid prototyping and evaluation of speech workloads.

Answer

A recording studio application for producing AI-generated music and audio

Answer

A professional audio editing tool for removing background noise from recordings

Answer

An IDE extension that adds speech commands to control code editors

Question 12

A data scientist is training a binary classification model to detect fraudulent transactions. The dataset contains only 1% fraudulent transactions. The model achieves 99% accuracy on the test set, but when deployed, it fails to detect most actual fraud cases. Which metric would best reveal this issue?

Accepted Answer

Recall. Recall (sensitivity) measures the proportion of actual positive cases correctly identified. In this highly imbalanced dataset (1% fraud), a model can achieve 99% accuracy by simply predicting 'non-fraud' for every transaction, which yields zero true positives. Recall reveals this failure because it focuses solely on how many fraudulent transactions were caught, ignoring the vast majority of non-fraud cases.

Answer

Accuracy

Answer

Precision

Answer

F1 score

Question 13

A company uses a generative AI model to create blog posts. They want to ensure that the model's output never contains offensive or harmful language before the content is published. They implement a system that checks the generated text against a list of prohibited terms and blocks or edits the content if necessary. Which type of safety measure is this?

Accepted Answer

Post-processing content filtering. The described system operates after the model generates text, scanning the output against a prohibited terms list and blocking or editing it. This is a classic post-processing content filtering approach, distinct from modifying the model's training data, prompts, or weights. Azure AI Content Safety is an example of such a post-processing filter that can be applied to generative AI outputs.

Answer

Pre-training data cleaning

Answer

Prompt engineering with safety instructions

Answer

Model fine-tuning on safe examples

Question 14

What is 'question answering' in Azure AI Language and what are its two main types?

Accepted Answer

Custom QA (trained on your documents) and prebuilt QA (document provided at query time). Azure AI Language's 'question answering' feature provides two distinct capabilities: Custom QA, where you train a model on your own documents (e.g., PDFs, FAQs) to answer questions from that knowledge base, and Prebuilt QA, which uses a document provided at query time to extract answers without prior training. This distinction is fundamental to how the service is deployed—either as a persistent, trained knowledge base or as an on-the-fly extraction from a user-supplied document.

Answer

Multiple-choice question generation and open-ended answer scoring

Answer

Structured QA for databases and unstructured QA for text documents

Answer

Real-time QA for chatbots and batch QA for scheduled document processing

Question 15

A data scientist trains a binary classification model to detect spam emails. The dataset contains 95% legitimate emails (negative class) and 5% spam (positive class). The model predicts all emails as legitimate. The accuracy is 95%, but the model is useless. Which metric would best indicate the model's failure?

Accepted Answer

Recall. Recall (sensitivity) measures the proportion of actual positive cases correctly identified. With 5% spam and the model predicting all as legitimate, recall is 0% because no spam emails are detected. This directly exposes the model's failure to identify the positive class despite high accuracy.

Answer

Precision

Answer

F1 score

Answer

Specificity