AB-731: AI Transformation LeaderDemo

A pretrained generative AI model is trained on a large, broad, general-purpose dataset to learn general patterns, language structure, and knowledge. Fine-tuning takes that pretrained model and continues training it on a smaller, narrower, domain- or task-specific dataset so it performs better on specialized tasks. This is why option C is correct: it accurately reflects that pretraining establishes broad general capability while fine-tuning adapts the model to a specific domain or use case.

Azure AI Search is designed to index and query large volumes of content, enabling users to retrieve relevant information using natural language queries. In generative AI solutions, it commonly serves as the retrieval component in Retrieval-Augmented Generation (RAG) patterns, allowing the AI system to search across large collections of documents and reports and return the most relevant results based on natural language input, which can then be used to ground AI-generated responses in organizational data.

Anonymizing the source data to strip out Personally Identifiable Information before it is processed by the AI solution directly reduces privacy risk at its source, since the system can no longer expose or misuse identifiable customer data while still allowing the underlying feedback content to be analyzed for product design insights. This aligns with core privacy-by-design and data minimization principles (e.g., GDPR data minimization requirements and Microsoft's Responsible AI "privacy and security" pillar), which call for reducing exposure of personal data as early as possible in a system's lifecycle rather than relying on later remediation such as deletion after the fact, keeping data attributable (which increases rather than decreases exposure), or deferring privacy reviews to the end of the project.