Generative AI Models and Selection — Azure AI Fundamentals AI-901 practice questions

A large language model (LLM) is best described as:

1. A managed relational database engine for structured rows
2. A foundation model trained on text that generates language
3. A network protocol for passing messages between services
4. A web application framework for building user interfaces

Answer: B — A foundation model trained on text that generates language

B) Right answer - an LLM is a foundation model with a very large parameter count trained on huge text corpora, and it generates human-like language. A) A database stores and queries records rather than generating text. C) A protocol defines message exchange on the wire. D) A framework structures application code and has no learned parameters.

An application sets a low max tokens value and users report that answers stop mid-sentence. What does this parameter actually do?

1. It instructs the model to write more concisely so answers fit within a budget
2. It sets a hard ceiling on the number of tokens generated
3. It limits how many tokens of prompt the deployment will accept per request
4. It reserves a token allowance that is refunded when a response finishes early

Answer: B — It sets a hard ceiling on the number of tokens generated

B) Correct — max tokens is a hard cap on generation, and when it is hit the response simply stops, which is why answers appear truncated. A) Tempts as the single most common misconception; the parameter does not change the model's style, only where generation halts. C) Tempts by reversing the direction; the cap applies to output, while input is bounded by the context window. D) Tempts by inventing a billing mechanic that does not exist.

An engineering team needs a model that works through multi-step derivations and self-checks intermediate conclusions on hard planning problems, accepting slower responses in exchange. Which selection is MOST appropriate?

1. A small, low-latency chat model tuned for high-volume conversational replies
2. A reasoning model that deliberates further before answering
3. An embedding model, then rank candidate plans by their vector similarity
4. A text model with temperature raised so it explores more solution paths

Answer: B — A reasoning model that deliberates further before answering

B) Correct — reasoning models are selected precisely when a task benefits from extended internal deliberation before responding, trading latency and cost for accuracy on complex problems. A) Tempts because small models are attractive operationally, but the scenario explicitly accepts latency for depth. C) Tempts because similarity ranking sounds like evaluation, but embeddings measure closeness of meaning, not the validity of a derivation. D) Tempts as a common misconception that higher temperature produces better reasoning; it produces more varied wording, not more rigour.

A team must decide between grounding with retrieval and fine-tuning. Their content is proprietary, changes weekly, and answers must cite the exact source paragraph. Which choice is BEST and why?

1. Fine-tuning, because proprietary content should be absorbed into the model's own weights
2. Fine-tuning, because a trained model responds faster than one that performs a lookup first
3. Grounding, because retrieved passages stay current and can be cited in the answer
4. Grounding, because it removes the need to write a system prompt for the assistant

Answer: C — Grounding, because retrieved passages stay current and can be cited in the answer

C) Correct — retrieval keeps pace with weekly changes without retraining and makes citation straightforward because the source passage is already in the request. A) Tempts because folding knowledge into weights sounds definitive, but it re-freezes the content and offers no citation path. B) Tempts with a true statement about latency that does not outweigh the stated currency and citation requirements. D) Tempts because it reaches the right conclusion, but the stated reason is wrong: grounding does not replace the system prompt.

Which is NOT a typical use case for generative AI?

1. Drafting first-pass marketing copy
2. Suggesting code completions in an editor
3. Producing exact payroll tax figures
4. Summarizing a long meeting transcript

Answer: C — Producing exact payroll tax figures

C) Correct — payroll figures must be exact and reproducible, so deterministic calculation, not a probabilistic model, is the right tool. A, B and D) Are all standard generative workloads where fluent draft output is useful.

Why is an embedding model not a chat model, and what is it for?

An embedding model converts text into a numeric vector that captures meaning; it generates no prose. Its job is similarity search — storing vectors for your content so a query can retrieve the closest passages, which are then handed to a chat model as grounding.

What does the max tokens setting limit, and what is the classic mistake?

It caps the length of the generated response, not the length of the prompt. The classic mistake is setting it too low and getting output cut off mid-sentence, then blaming the model — the response is truncated, not finished.

What is Generative AI?

AI that CREATES new content — text, images, code, audio — based on patterns learned from training data.

What is a model deployment, and why is it separate from the model itself?

A deployment is a named, addressable instance of a catalog model within your project, with its own capacity, content filter and access controls. Applications call the deployment name, so you can swap or upgrade the underlying model version without changing client code.