AB-731 sample exam questions

Grounding anchors a model's responses in verified, trusted data sources such as company documents, which is what D describes. Language support, usage metrics and natural language interaction are unrelated to tying output to authoritative data.

Generative AI learns patterns from large datasets and uses them to produce new content such as text, images, or code. Predictive AI forecasts future outcomes from historical data, and prescriptive AI recommends actions, so neither of them creates original content.

Blueprints combine drawings and text and require summarizing, question answering and extraction, so a multi-modal model that processes both images and text fits. OCR only extracts text, text completion handles text only, and summarization alone does not cover images or extraction.

Azure Machine Learning delivers strategic value by building predictive models on historical data, such as demand forecasting. Digitizing paper, spreadsheet entry and rule-based location emails are simple automation tasks that do not need custom machine learning.

Trust concerns about security and accuracy commonly slow adoption, and poor data quality undermines results because AI depends on its data. Involving business and compliance stakeholders from the start aligns projects with real goals and regulatory requirements, which improves the chances of success.

Foundry is the pro-code Azure platform for developers and data professionals building custom AI solutions. Low-code makers and business users are better served by Copilot Studio. Foundry access depends on an Azure subscription and Azure roles, so no Microsoft 365 Copilot license is required.

Azure Machine Learning is used to train models on historical data to find patterns and predict future outcomes. Document Intelligence and Content Understanding extract data from documents, and Foundry is a generative AI platform rather than the predictive modeling tool asked for.

Switching to a smaller or cheaper model and optimizing prompts to use fewer tokens both cut cost with little quality loss. Content moderation adds a step, fine-tuning adds cost, and a prebuilt model on standard billing is not charged by hosting hours.

Azure Language in Foundry Tools provides named entity recognition, which identifies people, locations and organizations in text. Vision works on images, Content Safety detects harmful content, and Speech handles audio, so none extract entities from article text.

Microsoft Foundry, formerly Azure AI Foundry, is the unified Azure platform for building, customizing, evaluating and managing AI apps, agents and models. Power Platform targets low-code business apps, and Azure Machine Learning focuses on training and operating custom machine learning models rather than the full AI app lifecycle.

An AI council gives cross-business guidance, oversight and coordination, keeping projects aligned with strategy and responsible AI principles. Letting departments act alone causes inconsistency, and delegating to IT or outsourcing removes business ownership of strategy and accountability.

Declarative agents that access work data in Microsoft 365 require the Microsoft 365 Copilot add-on license for the user. Copilot Chat alone, with pay-as-you-go, can use agents but without grounding in Microsoft Graph work data, and a Copilot Studio license is not the requirement.

Governance needs clear accountability norms across business and technical teams and a review process that checks initiatives for responsible AI alignment. Letting departments diverge, leaving ownership to engineers, or covering only regulated data leaves gaps in consistent, organization-wide governance.

Azure Machine Learning builds predictive models, so predicting customer retention from historical data is its use case. Creating descriptions from images, generating campaigns and summarizing transcripts are generative AI tasks better handled by foundation models.

Least privilege limits access to only what each user needs, and Copilot honors those existing permissions when retrieving data, which makes it a data security control. Classification labels content, compliance addresses regulatory obligations, and data loss prevention stops sensitive information from being shared or leaked.