OCI 1Z0-1122-25 Sample Questions with Explanations

These original sample questions are designed to help you check how the exam topics appear in decision-style prompts. They are not taken from the live exam.

Use these sample questions as a guided self-assessment for OCI AI Foundations Associate (1Z0-1122-25) topics such as AI workload recognition, training versus inference, evaluation metrics, responsible AI, grounding, data quality, and service-fit decisions. The prompts are fundamentals-oriented, but still require judgment.

Where these questions fit in the 1Z0-1122-25 guide

The sample set below is part of the Oracle OCI 1Z0-1122-25 guide path:

• Oracle certification hub
• OCI guides
• 1Z0-1122-25 exam guide
• 1Z0-1122-25 cheat sheet
• 1Z0-1122-25 study plan
• 1Z0-1122-25 FAQ
• 1Z0-1122-25 resources
• 1Z0-1122-25 glossary

1Z0-1122-25 OCI AI Foundations sample questions

Work through each prompt before opening the explanation. For foundations questions, identify the AI lifecycle stage and risk before choosing a tool or technique.

Question 1

Topic: Choosing the evaluation signal

A team trains a model that performs very well on training data but poorly on new customer records. Which issue best explains the pattern?

• A. Overfitting, because the model learned training-specific patterns that do not generalize.
• B. Inference, because any prediction on new data is always less accurate.
• C. Grounding, because all models need retrieval documents before they can classify records.
• D. Tokenization, because structured customer records cannot be used in machine learning.

Best answer: A

Explanation: High training performance with weak performance on unseen data is the classic overfitting signal. The right response is to improve generalization, validation, data handling, or model complexity rather than celebrating training accuracy.

Why the other choices are weaker:

• B describes using a model, not the reason for the performance gap.
• C over-applies grounding, which is mainly a generative/RAG concept.
• D is false; structured data is common in ML workflows.

What this tests: Model evaluation, overfitting, and training-versus-validation reasoning.

Related topics: Overfitting; Evaluation; Generalization; Model lifecycle

Question 2

Topic: Responsible AI control choice

A business wants to deploy an AI feature that summarizes customer support cases. The cases may contain sensitive customer information. What should be addressed before broad rollout?

• A. Only the size of the model, because larger models eliminate privacy risk.
• B. Data handling, access controls, output review, and monitoring for unsafe or inaccurate summaries.
• C. Only prompt wording, because responsible AI is mainly a style-guide issue.
• D. Removing all human review, because automation is the primary goal of AI.

Best answer: B

Explanation: Responsible AI includes data protection, access boundaries, quality checks, monitoring, and human oversight where risk requires it. Sensitive support data makes governance part of the design, not an afterthought.

Why the other choices are weaker:

• A confuses model capability with privacy and governance.
• C reduces responsible AI to wording and ignores controls.
• D removes oversight in a sensitive workflow.

What this tests: Responsible AI, privacy, access controls, monitoring, and deployment judgment.

Related topics: Responsible AI; Privacy; Monitoring; Governance

Question 3

Topic: Training versus inference

A deployed model receives a new support ticket and returns a category label for routing. Which lifecycle stage is this?

• A. Training, because the model sees a new example.
• B. Inference, because the trained model is being used to produce an output for new input.
• C. Data labeling, because every categorized ticket changes the training set automatically.
• D. Feature engineering, because the category label is the same as a feature.

Best answer: B

Explanation: Inference is the use of a trained model to generate a prediction, classification, or output for new input. Seeing a new ticket does not automatically mean the model is being retrained.

Why the other choices are weaker:

• A confuses using a model with updating model parameters.
• C assumes an automatic labeling pipeline that is not described.
• D confuses input features with output labels.

What this tests: AI lifecycle vocabulary and distinguishing training from deployed use.

Related topics: Inference; Training; Classification; AI lifecycle

Independent study note

Tech Exam Lexicon and IT Mastery are independent study tools. They are not affiliated with, endorsed by, or sponsored by Oracle or any certification body.