AI-900 Practice Question: Describe features of generative AI workloads on Azure
A developer uses Azure OpenAI Service to generate product descriptions. They want to ensure that the model only considers the most likely tokens that together have a cumulative probability of 0.95, ignoring very low-probability tokens that could lead to nonsensical outputs. Which parameter should they configure?
⚠ Common exam trap
Candidates often confuse Top_p with Temperature, assuming both control randomness, but Temperature scales logits without filtering low-probability tokens, whereas Top_p directly removes them based on cumulative probability mass.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Top_p
(Top_p) is correct because the developer wants to limit token selection to those with a cumulative probability of 0.95, which is exactly what the Top_p (nucleus sampling) parameter controls. By setting Top_p to 0.95, the model will only consider the smallest set of tokens whose combined probability mass reaches 0.95, effectively ignoring low-probability tokens that could produce nonsensical outputs.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Temperature
Why it's wrong here
Temperature scales the logits before the softmax is applied, changing how sharply or evenly the probability distribution is shaped. A low temperature (e.g., 0.2) makes high-scoring tokens dominate, while a high temperature (e.g., 0.8 or 1.5) makes all tokens more likely, including improbable ones. Crucially, it never removes a token from the sampling set; every vocabulary token remains eligible, so temperature is not a cumulative probability cutoff.
When this WOULD be correct
A developer wants to control the creativity of generated text by adjusting the randomness of token selection. For example, a lower temperature (e.g., 0.2) makes the model more deterministic, while a higher temperature (e.g., 0.8) increases diversity. The question would ask: 'Which parameter adjusts the randomness of the model's output?'
- ✓
Top_p
Why this is correct
Top_p (nucleus sampling) restricts sampling to the smallest set of tokens whose cumulative probability mass reaches the specified p value. If p is 0.9, only tokens that together account for 90% of the probability distribution are considered, and all lower-probability tail tokens are pruned. This dynamically narrows the candidate vocabulary based on the current context, producing more focused and coherent completions.
- ✗
Frequency penalty
Why it's wrong here
The frequency penalty reduces a token's score proportionally to how many times that token has already been generated, making repeated tokens progressively less likely as text is produced. It is a per-occurrence penalty applied to the final logits and depends on the current token counts, not on the cumulative probability mass of candidate tokens. This mechanism addresses repetition, not nucleus-style probability thresholding, so it cannot serve the same role as top_p.
When this WOULD be correct
A developer wants to reduce repetitive word usage in generated text, such as avoiding the same product feature being mentioned multiple times in a description. The frequency penalty parameter would be configured to decrease the likelihood of tokens that have already appeared frequently.
- ✗
Presence penalty
Why it's wrong here
The presence penalty applies a fixed additive adjustment to the logits of any token that has already appeared anywhere in the generated sequence, encouraging the model to introduce new topics and avoid repetition. It acts after the model computes token scores, not by filtering candidate tokens based on cumulative probability. Therefore, it does not define a nucleus threshold and cannot constrain the token pool the way top_p does.
When this WOULD be correct
A scenario where the question asks: 'A developer wants to reduce the repetition of words or phrases that have already been generated in a text completion. Which parameter should they adjust?'
Option-by-option analysis
Why each answer is right or wrong
Understanding why wrong answers are wrong — and when they would be correct — is what separates a 750 score from a 900. The AI-900 exam frequently reuses these exact scenarios with slightly different constraints.
✓Top_pCorrect answer▾
Why this is correct
Top_p (nucleus sampling) restricts sampling to the smallest set of tokens whose cumulative probability mass reaches the specified p value. If p is 0.9, only tokens that together account for 90% of the probability distribution are considered, and all lower-probability tail tokens are pruned. This dynamically narrows the candidate vocabulary based on the current context, producing more focused and coherent completions.
✗TemperatureWrong answer — click to see why▾
Why this is wrong here
Temperature controls the randomness of token selection by scaling logits, not by filtering a cumulative probability mass. The question specifies selecting tokens with a cumulative probability of 0.95, which is exactly what top_p (nucleus sampling) does.
★ When this WOULD be the correct answer
A developer wants to control the creativity of generated text by adjusting the randomness of token selection. For example, a lower temperature (e.g., 0.2) makes the model more deterministic, while a higher temperature (e.g., 0.8) increases diversity. The question would ask: 'Which parameter adjusts the randomness of the model's output?'
Why candidates choose this
Candidates often confuse temperature with top_p because both influence output diversity. They may think temperature also controls a probability threshold, but temperature scales logits before softmax, whereas top_p directly filters by cumulative probability.
✗Frequency penaltyWrong answer — click to see why▾
Why this is wrong here
The frequency penalty reduces repetition by penalizing tokens based on their existing frequency in the generated text, not by controlling the cumulative probability of token selection. It does not filter out low-probability tokens to achieve a target cumulative probability.
★ When this WOULD be the correct answer
A developer wants to reduce repetitive word usage in generated text, such as avoiding the same product feature being mentioned multiple times in a description. The frequency penalty parameter would be configured to decrease the likelihood of tokens that have already appeared frequently.
Why candidates choose this
Candidates may confuse frequency penalty with probability-based sampling, thinking that penalizing frequent tokens also helps avoid nonsensical low-probability outputs, but the mechanism is different.
✗Presence penaltyWrong answer — click to see why▾
Why this is wrong here
Presence penalty reduces the likelihood of repeating tokens that have already appeared, but it does not control the cumulative probability mass of token selection; the described behavior (considering only tokens with cumulative probability 0.95) is controlled by top_p sampling.
★ When this WOULD be the correct answer
A scenario where the question asks: 'A developer wants to reduce the repetition of words or phrases that have already been generated in a text completion. Which parameter should they adjust?'
Why candidates choose this
Candidates may confuse presence penalty with probability-based filtering because both involve modifying token selection, but presence penalty specifically targets repetition rather than cumulative probability thresholds.
Analysis generated from the official AI-900blueprint and verified against question context. The “when correct” sections are what AI assistants cite when candidates ask “what’s the difference between these options?”
Go deeper
Related to this question
Learn chapter
Azure Machine Learning Studio
Key term
Model
In IT and AI, a model is a trained mathematical representation that learns patterns from data to make predictions or decisions.
Key term
Azure OpenAI Service
Azure OpenAI Service is a cloud platform from Microsoft that lets developers use powerful artificial intelligence models, like GPT-4, to build applications that can understand and generate human-like text, code, images, and more.
About these practice questions
One of 985 original AI-900 practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. Learn why practice questions differ from exam dumps →
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This AI-900 practice question is part of Courseiva's free Microsoft certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the AI-900 exam.