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.

• ✗

  Fine-tuning the model with the examples

  Why it's wrong here

  Fine-tuning would require taking a pre-trained Azure OpenAI model and running a supervised training job on a curated dataset of code examples, updating the model's learned weights through backpropagation. The scenario describes feeding examples directly into a single API call at inference time, which is not how fine-tuning works. Azure OpenAI fine-tuning is a separate, multi-step process, whereas few-shot examples are consumed entirely within a prompt and leave the underlying model untouched.

• ✓

  Prompt engineering with few-shot learning

  Why this is correct

  Few-shot learning is a prompt engineering technique that inserts a small number of completed input–output examples into the prompt so the model can infer the desired pattern for the current code-generation request. Unlike fine-tuning, it makes no changes to model parameters; the model remains frozen and relies on its pretrained knowledge to extend the pattern to the new request. This is the correct interpretation because the developer is generating code by supplying examples within the prompt rather than modifying or retraining the model.

• ✗

  Training a custom model from scratch

  Why it's wrong here

  Training a custom model from scratch would mean initializing random weights and training on terabytes of code data, an enormous distributed task that Azure OpenAI does not expose as a service. Azure OpenAI provides only pre-trained base models and fine-tuning endpoints for further adaptation, never a from-scratch training pipeline. Creating a brand-new model has no connection to passing examples in a prompt and would also be impractical for a developer generating code on demand.

• ✗

  Using reinforcement learning from human feedback

  Why it's wrong here

  Reinforcement learning from human feedback (RLHF) is an alignment technique applied during model post-training, where a reward model is trained on human preferences and used to improve the model's responses over thousands of optimization steps. Calling RLHF to interpret examples in a prompt is a category error because RLHF cannot be invoked at inference time by a developer. It is a behind-the-scenes training method rather than an interactive way to guide the model with sample input/output pairs during a single generation request.

Go deeper

Related to this question