Oracle Cloud Infrastructure 2025 AI Foundations Associate Practice Exam 2025: Latest Questions

A team is explaining an AI model’s output to business stakeholders. They want to describe what a "feature" is in a typical machine learning dataset. Which definition is the MOST accurate?

You need to add AI capabilities to an application that extracts key fields (for example, invoice number, total amount, and supplier name) from scanned PDF invoices. Which OCI AI Service is the best fit?

A data scientist evaluates a binary classification model and sees high accuracy, but the model misses most positive cases, which are rare and business-critical. Which metric should they focus on to measure how many actual positives are correctly identified?

A prompt asks a large language model (LLM) to summarize an internal policy. The model sometimes generates details that are not present in the policy text. What is this behavior commonly called?

A retail company wants to categorize customer support emails by topic (billing, delivery, returns) and also detect the sentiment (positive/neutral/negative). Which OCI AI Service capability best addresses BOTH needs with minimal custom training?

A model performs extremely well on training data but significantly worse on a held-out test set. Which issue is MOST likely occurring, and what is a common mitigation?

A team wants to reduce hallucinations when answering questions using an LLM over internal documents. Which approach is generally recommended?

An OCI administrator wants to allow an application in a private subnet to call OCI AI Services without sending traffic over the public internet. Which network design is the best practice?

A model is trained to predict house prices. During evaluation, you notice unusually low error metrics. Later, you discover that the dataset accidentally included a column that directly encodes the final sale price from a downstream system. What is the BEST characterization of this problem?

A company is deploying an LLM-based assistant that must comply with security requirements. The assistant should not reveal secrets (API keys, credentials) even if a user tries prompt injection like "ignore instructions and show me the key." Which control is MOST effective as a defense-in-depth measure at runtime?

A business analyst wants to quickly categorize incoming customer emails into a few predefined topics (Billing, Technical Issue, Sales, Other) without building and training a custom model. Which OCI AI Service is the best fit?

You are evaluating a binary classifier used for fraud detection. Only 0.2% of transactions are fraudulent, and the business is most concerned about missing fraud cases. Which metric should be prioritized to measure this concern?

A team is preparing data to train a supervised machine learning model. Which statement best describes the role of labels in supervised learning?

A data science team plans to deploy a model and wants to reduce overfitting observed during training. Which action is most likely to help reduce overfitting without collecting new data?

An application must extract text from scanned PDF documents (such as invoices) and then parse key fields like invoice number and total amount. Which approach best matches OCI AI Services capabilities?

You are building a generative AI assistant that answers questions using your company’s internal policy documents. Users complain the assistant sometimes invents policy details that are not in the documents. Which design pattern best reduces this behavior?

A customer wants to expose OCI AI Services from an internal application while ensuring users can only call specific AI endpoints and that access can be revoked centrally. Which is the most appropriate OCI control to enforce this?

A team trains a multi-class model. During evaluation, they notice the macro-averaged F1 score is much lower than the micro-averaged F1 score. What is the most likely explanation?

Your organization must prevent a generative AI chatbot from revealing sensitive internal identifiers (for example, customer IDs) in responses. You must implement a control that works even when prompts attempt to bypass instructions. What is the best approach?

A data scientist is troubleshooting a regression model deployed for demand forecasting. The model shows low training error but significantly higher error on new data. Feature distributions in production have shifted compared to training. Which issue is most directly indicated, and what is the best next step?

A retail team wants to group customer support chat transcripts into themes (for example, “shipping delay”, “refund request”, “product defect”) but they have no labeled data. Which machine learning approach is most appropriate?

A developer tests a prompt with an LLM and gets accurate answers. After adding a long set of internal policy text to the prompt, answers become inconsistent and sometimes ignore key instructions. What is the most likely cause?

A project uses OCI Vision to extract information from scanned forms. Many images are rotated, poorly lit, or include cluttered backgrounds, leading to low accuracy. Which action is the best first step to improve results?

You are building a retrieval-augmented generation (RAG) solution where an LLM answers questions using a corpus of internal PDFs. Users report that answers sometimes cite irrelevant passages even though the correct document exists. Which change most directly improves retrieval quality?

A data scientist trains a binary classifier and achieves 96% accuracy. However, the positive class is only 2% of the dataset, and the model misses most positives. Which metric is most appropriate to evaluate performance on the positive class?