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.

• ✗

  A configuration file for training AI models

  Why it's wrong here

  A configuration file is a static artifact—usually YAML or JSON—that specifies model architecture, hyperparameters, dataset paths, and training procedures such as learning rate, batch size, and epoch count. It is consumed by a training pipeline to set up an experiment, not by the model at inference time. A prompt, by contrast, is dynamic, user-supplied text provided when the model is asked to generate a response, so it cannot be a configuration file.

• ✓

  The input text or instruction given to a generative AI model to guide its output

  Why this is correct

  A prompt is the natural-language instruction or query that a user supplies to a generative AI model (for example, GPT, Claude, or DALL·E) to condition the output. The model treats the prompt as its starting context and autoregressively generates a continuation or image based on that text. Because the generated content directly follows from the prompt's wording, specificity, and framing, prompt design has a direct impact on output quality, beyond any fixed model weights.

• ✗

  A reward signal used in reinforcement learning

  Why it's wrong here

  Reinforcement learning (RL) trains an agent by pairing actions with a scalar reward signal—a numeric feedback value that tells the agent how good its action was in a given state. The agent's goal is to maximize cumulative reward over time, not to consume a text instruction as input. This is fundamentally different from a prompt, which is a linguistic specification used only for generative inference; RL does not require a prompt at all. Even in RLHF, the human preference model supplies rewards, but the prompt is the initial text given to the language model, not the reward.

• ✗

  A type of neural network activation function

  Why it's wrong here

  Activation functions such as ReLU, sigmoid, or tanh are mathematical transformations applied to the weighted sum of a neuron's inputs to introduce non-linearity into the network's output. They are deterministic, defined once as part of the network architecture, and applied identically during forward propagation. A prompt is not a mathematical function; it is an external input sample—variable-length text that is tokenized and embedded before the model processes it, so it belongs to a different layer of the system.

Go deeper

Related to this question