UUS Practice Tests⌕ Search
IT Certifications · 1Z0-1122-26

Oracle Cloud Infrastructure AI Foundations Associate Exam Practice Test

The Oracle Cloud Infrastructure AI Foundations Associate Exam practice page covers introductory artificial intelligence and machine learning concepts alongside examples of AI services in OCI. Practice questions address topics such as neural-network training, dataset preparation, generative language model settings, responsible AI decisions, and selecting services for speech transcription or invoice data extraction.

Exam overview

Exam providerOracle
Exam code1Z0-1122-26
CategoryIT Certifications

About the Oracle Cloud Infrastructure AI Foundations Associate Exam

Core concepts include supervised learning, reinforcement learning, neural-network backpropagation, and the preparation of reliable training and test data. Questions also explore practical choices such as converting measurements to consistent units, checking for duplicate records before splitting datasets, and tracking model versions and configurations. Generative AI topics include how temperature affects language-model output and how prompt injection can exploit user-provided content. OCI service scenarios involve matching a task to a capability, including converting speech to text or extracting fields from documents. Responsible-use questions consider safeguards when AI contributes to decisions affecting people. Together, these areas connect foundational AI terminology with the judgment needed to interpret common model behaviors, identify data-quality risks, and choose an appropriate cloud AI capability for a described task.

Sample Practice Questions

Review a selection from this exam's published US Practice Tests practice bank.

Question 1

In a neural network, what does backpropagation primarily compute during training?

  1. AGradients that indicate how model parameters affect the loss
  2. BA random ordering of the evaluation examples
  3. CThe number of features to retain in the dataset
  4. DThe final predictions for future production requests
Show answerCorrect answer: A

Backpropagation applies the chain rule to calculate gradients of the loss with respect to model parameters. An optimizer can use those gradients to update the parameters.

Question 2

What is a prompt injection risk in an application that uses a language model to process user-provided content?

  1. AThe input may contain instructions intended to override the application's intended rules
  2. BThe model's training data is automatically deleted after each response
  3. CThe application is guaranteed to return a response in JSON
  4. DThe model converts every input into an image
Show answerCorrect answer: A

Prompt injection involves untrusted content attempting to manipulate a model's behavior or bypass intended instructions. Applications should treat user content as untrusted and apply suitable safeguards.

Question 3

Why is it useful to keep a record of a model version and its configuration when deploying an AI application?

  1. ATo support traceability and investigation of behavior changes
  2. BTo guarantee that the model never makes an error
  3. CTo remove the need for access controls
  4. DTo ensure every request uses identical input data
Show answerCorrect answer: A

Recording versions and configurations helps teams trace which system produced an output and investigate changes after updates. It cannot guarantee correctness or replace security controls.

Question 4

Which practice is most appropriate when using an AI system to help make decisions that affect individuals?

  1. AReview outputs for potential bias and keep appropriate human oversight
  2. BAssume the model is neutral because it is automated
  3. CUse model outputs without documenting how they are applied
  4. DRemove all evaluation steps once the model is deployed
Show answerCorrect answer: A

AI systems can reflect biases in their data or design. Assessing outcomes and retaining suitable human oversight are responsible practices, particularly when decisions may significantly affect people.

Question 5

In a generative language model, what does temperature generally influence during text generation?

  1. AThe amount of training data stored in the model
  2. BThe randomness or diversity of token selection
  3. CThe maximum number of users allowed to access a service
  4. DThe language in which the model was originally trained
Show answerCorrect answer: B

Temperature adjusts the sharpness of the model's token-selection distribution. Lower values tend to produce more predictable choices, while higher values can increase variation.

Question 6

When preparing a dataset for machine learning, why should duplicate or near-duplicate records be checked before splitting data into training and test sets?

  1. AThey can leak highly similar examples across the splits and make evaluation look overly optimistic
  2. BThey automatically increase the number of distinct real-world cases
  3. CThey force all models to use classification rather than regression
  4. DThey prevent the training data from being stored in the cloud
Show answerCorrect answer: A

If duplicates or very similar records appear in both training and test sets, a model may effectively encounter test examples during training. This can inflate evaluation results and provide a misleading estimate of performance on genuinely new data.

Question 7

A dataset contains measurements recorded in centimeters, but a model requires values in meters. What is the direct purpose of converting the measurements?

  1. AData normalization by changing measurement units
  2. BData augmentation by creating new examples
  3. CFeature selection by removing irrelevant variables
  4. DData labeling by assigning target values
Show answerCorrect answer: A

Converting centimeters to meters changes the numerical representation while preserving the underlying measurement. It does not create examples, select features, or assign labels.

How to use this practice test

Access the practice questions for free. Each session presents up to 20 random questions from the available bank; results appear after the session. Use missed or uncertain answers to identify concepts to revisit, such as data leakage risks from duplicate records, prompt injection, model tracking, or the distinction between speech recognition and document extraction.

Preparation tips

  • Explain what backpropagation computes and how it supports neural-network training; distinguish this from reinforcement learning based on feedback or rewards.
  • Work through dataset-preparation examples involving unit conversion, duplicate detection, and separating training data from test data without contaminating evaluation.
  • Compare language-model temperature with prompt injection: one affects generation variability, while the other concerns unsafe instruction handling in user-provided content.
  • Practice mapping OCI task descriptions to capabilities, especially speech-to-text transcription and invoice field extraction with document understanding.
  • For responsible AI scenarios, identify how human oversight, impact on affected individuals, and traceability through model-version records can inform safer decisions.
Official sourceOfficial exam page ↗