AI-900 Practice Question: Describe features of generative AI workloads on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Connecting the model to electrical ground to prevent static during training

  Why it's wrong here

  This option is a pun on the word 'grounding.' In electrical engineering, grounding connects a circuit to earth to prevent static buildup, but in generative AI, grounding refers to anchoring model responses to factual, retrieved knowledge sources. There is no literal electrical connection involved in model training or inference; instead, grounded models are constrained by evidence inserted into the prompt. Treating the term literally reflects a fundamental misunderstanding of AI terminology.

• ✓

  Anchoring model responses to specific, retrieved source documents to improve factual accuracy

  Why this is correct

  Grounding anchors a model's output by injecting retrieved, authoritative source passages into the prompt context, forcing the model to generate responses that are consistent with verified evidence. This directly reduces hallucination and improves factual accuracy, because the model can cite and reason over the supplied documents rather than relying solely on parametric memory. It is the core mechanism behind retrieval-augmented generation (RAG) systems.

• ✗

  The process of converting floating-point weights to integer values for deployment

  Why it's wrong here

  Converting floating-point weights to integer values is quantization, a model optimization technique that reduces memory footprint and speeds up inference by representing weights as lower-precision integers such as INT8. Quantization is entirely separate from grounding: it changes how the model is stored and executed, not how its outputs are anchored to external knowledge sources. While quantization can make deployment cheaper, it does nothing to improve factual grounding or source attribution.

• ✗

  Setting the baseline performance metrics before model fine-tuning begins

  Why it's wrong here

  Setting baseline performance metrics before fine-tuning is an evaluation practice, not a grounding mechanism. A baseline—such as accuracy, F1, or BLEU score—provides a reference point to measure whether fine-tuning improved the model, but it does not connect the model to a knowledge base or source documents. Grounding, by contrast, is a runtime prompting and retrieval technique that controls the source of information the model uses when generating an answer.

Go deeper

Related to this question