Question bank
Oracle Cloud Infrastructure 2026 AI Foundations Associate
40 Oracle Cloud Infrastructure 2026 AI Foundations Associate questions with the correct answer and a written explanation, organised across 6 exam domains.
- 40 questions
- 6 domains
- Foundation level
certilumQuestion index
Jump straight to any question. Answers and explanations are inline.
AI Foundations
5 questions in this domain
- Question 1Single Choice
At a foundational level, what capability is artificial intelligence designed to simulate in computer systems?
- AHuman-like reasoning and learningCorrect answer
- BDatabase query optimization
- CMotherboard voltage control
- DNetwork address translation
Why this is the answer
AI is designed to allow machines to perform human-like mental tasks such as learning, reasoning, and problem solving.
- Question 2Multiple Select
Which three disciplines, from the choices below, are widely recognized as core subfields of artificial intelligence? (Choose three.)
- AMachine learningCorrect answer
- BNatural language processingCorrect answer
- CDeep learningCorrect answer
- DSQL performance tuning
Why this is the answer
Machine learning, natural language processing, and deep learning are AI subfields. SQL performance tuning happens in databases and is not part of AI foundations.
- Question 3Interactive
Look at the code block and determine the type of machine learning task this linear formula is used to solve.
- ARegressionCorrect answer
- BClassification
- CClustering
- DReinforcement learning
Why this is the answer
The formula calculates a continuous numeric value from an input feature. Predicting continuous values is the defining task of regression.
- Question 4Single Choice
In supervised learning, what is the primary purpose of training a model with labeled data?
- AMap inputs to known correct outputsCorrect answer
- BRemove duplicate records from the dataset
- CConvert unstructured data to JSON
- DReduce network latency between users
Why this is the answer
Labeled data pairs inputs with expected outputs. Supervised learning uses those pairs to learn a mapping that can be applied to new data.
- Question 5Multiple Select
Which three use cases below are common applications of AI in everyday business environments? (Choose three.)
- ARecognizing objects in imagesCorrect answer
- BIdentifying sentiment in customer reviewsCorrect answer
- CProviding conversational support through chatbotsCorrect answer
- DCompiling source code into an executable program
Why this is the answer
Image recognition, sentiment analysis, and conversational chatbots are standard AI applications. Source-code compilation follows deterministic rules and is not considered AI.
Deep Learning Foundations
6 questions in this domain
- Question 6Single Choice
In deep learning, what is a perceptron and how does it process input signals to produce an output?
- AA single-layer network using a step activation functionCorrect answer
- BA recurrent cell that stores memory over time
- CA specialized loss function for regression tasks
- DA data augmentation technique for image datasets
Why this is the answer
A perceptron is a single-layer neural network that makes decisions by applying weights to inputs and passing the result through a step activation function.
- Question 7Multiple Select
Which of the following are core building blocks that are commonly found in a fully connected neural network?
- ANeurons with weighted connectionsCorrect answer
- BBias valuesCorrect answer
- CActivation functionsCorrect answer
- DTables of unstructured text
Why this is the answer
A neural network is composed of neurons, weighted connections, biases, and activation functions; unstructured text tables are not a required neural-network building block.
- Question 8Interactive
Look at the provided training snippet. Which line applies the computed gradients to update the model weights?
- Aprediction = model(example)
- Bloss = criterion(prediction, target)
- Closs.backward()
- Doptimizer.step()Correct answer
Why this is the answer
After loss.backward() computes gradients, optimizer.step() applies them to update the model parameters; the other lines compute predictions, loss, or gradients.
- Question 9Single Choice
What is the main purpose of an activation function in a deep learning model?
- AIntroduce non-linearityCorrect answer
- BStore training data
- CInitialize all weights
- DReplace the loss function
Why this is the answer
Activation functions add nonlinearity, enabling networks to learn complex patterns. Without them, repeated linear transformations would collapse into a single linear transformation.
- Question 10Multiple Select
During neural network training, which of the following statements are accurate?
- AAn epoch is one full pass over the training dataset.Correct answer
- BThe loss value indicates how far predictions are from targets.Correct answer
- CThe optimizer uses gradients to update the model's weights.Correct answer
- DDropout is used to increase the size of the training set.
Why this is the answer
One epoch means one complete pass through the dataset, loss measures prediction error, and optimizers use gradients. Dropout is a regularization technique, not a data augmentation method.
- Question 11Interactive
Examine the compiled model architecture in the code. What type of output does the final layer produce?
- AA probability between 0 and 1Correct answer
- BA categorical label string
- CA negative unbounded score
- DA fixed-size matrix image
Why this is the answer
The last Dense layer has one neuron with a sigmoid activation, so it outputs a single probability between 0 and 1, typically for binary classification.
Generative AI and Large Language Models
9 questions in this domain
- Question 12Single Choice
What is generative AI?
- AAI that creates new content based on patterns learned from training dataCorrect answer
- BAI that only classifies existing data into predefined categories
- CA database system designed to store large volumes of text
- DA programming language used to build neural networks
Why this is the answer
Generative AI learns statistical patterns from training examples and produces new outputs, such as text, images, or audio, that resemble those patterns.
- Question 13Multiple Select
Which statements about large language models (LLMs) are true? Select all that apply.
- AThey are trained on large text corporaCorrect answer
- BThey can perform language tasks such as summarization and translationCorrect answer
- CThey treat text as a sequence of tokensCorrect answer
- DThey only work with numeric tabular data
Why this is the answer
LLMs are trained on massive text data and treat text as tokens. They support tasks such as summarization and translation; they do not process only structured numbers.
- Question 14Interactive
Review the code in the code block. What input is passed as the prompt?
- AThe phrase 'Once upon a time'Correct answer
- BThe task label that identifies text generation
- CThe model name used by the pipeline
- DThe numeric value that limits output length
Why this is the answer
The code creates a text-generation pipeline and sends the variable named prompt to the model. Its value is the phrase 'Once upon a time', so that phrase is the prompt.
- Question 15Single Choice
What is a token in the context of an LLM?
- AThe smallest unit of text that the model processesCorrect answer
- BA complete paragraph sent to the model
- CThe final answer returned by the model
- DA hyperparameter that controls training speed
Why this is the answer
LLMs process text in small units called tokens. Tokens can represent words, parts of words, numbers, or punctuation rather than complete paragraphs.
- Question 16Multiple Select
Which statements about hallucinations in generated AI content are true? Select all that apply.
- AIt can present fabricated information with confidenceCorrect answer
- BIt can happen when the model lacks a reliable source to followCorrect answer
- CIt is always a sign of a network failure
- DIt can appear in fluent, grammatically correct textCorrect answer
Why this is the answer
Hallucination is the generation of plausible but false content. It often happens when the model lacks reliable grounding, and even fluent output can be factually incorrect.
- Question 17Interactive
In the code block, what is the user asking the model to do?
- AExplain tokens in beginner-friendly termsCorrect answer
- BAct as a helpful assistant
- CPrint the completion result
- DName the language model being used
Why this is the answer
The message with the user role contains 'Explain tokens to a beginner.' This is the instruction sent to the model, asking it to explain tokens in simple language.
- Question 18Single Choice
What is a foundation model?
- AA large model trained on broad data that can be adapted to many downstream tasksCorrect answer
- BA dedicated model that can perform only one fixed task
- CA model that stores every training example and retrieves it verbatim
- DA database index used for searching documents
Why this is the answer
Foundation models are large neural models pre-trained on diverse data. They can be fine-tuned or prompted for many downstream tasks.
- Question 19Multiple Select
Which tasks are examples of generative AI applications? Select all that apply.
- AWriting a short story from a one-sentence ideaCorrect answer
- BCondensing a long document into bullet pointsCorrect answer
- CGenerating code from a natural language descriptionCorrect answer
- DSorting a list of customer names alphabetically
Why this is the answer
Generative AI creates new content. Writing, summarizing, and generating code are generative tasks, while alphabetically sorting a list is a deterministic operation.
- Question 20Interactive
In the code block, which numeric value controls the randomness of the generated output?
- AThe value 0.8 assigned to the temperature settingCorrect answer
- BThe value 100 assigned to the maximum token limit
- CThe text of the request included as the prompt
- DThe client object used to call the generation service
Why this is the answer
The parameter temperature is set to 0.8. Temperature controls randomness: lower values produce more deterministic output, while higher values allow more variation.
Machine Learning Foundations
7 questions in this domain
- Question 21Single Choice
Which of the following best defines artificial intelligence?
- AA field that enables machines to simulate human-like reasoning and learningCorrect answer
- BA process used only to optimize database queries
- CA technology used to increase CPU clock speed
- DA method for copying files between cloud regions
Why this is the answer
AI focuses on machines simulating cognitive abilities such as reasoning and learning. It is not limited to databases, hardware speed, or file replication.
- Question 22Multiple Select
Which two of the following are recognized subfields of artificial intelligence?
- AMachine LearningCorrect answer
- BStructured Query Language
- CNatural Language ProcessingCorrect answer
- DData Center Cooling
Why this is the answer
Machine learning and natural language processing are core parts of AI. SQL and cooling are infrastructure-related topics, not AI subfields.
- Question 23Interactive
The code sample creates a logistic regression model, trains it, and uses it to predict labels. Which workflow is shown?
- ATraining a model followed by inference on new dataCorrect answer
- BUnsupervised clustering of customer accounts
- CReinforcement learning through reward signals
- DManual data cleaning before analysis
Why this is the answer
The fit operation trains the model, and the predict operation applies the trained model to new inputs. This sequence is training followed by inference.
- Question 24Single Choice
In supervised learning, what is the main purpose of labels in the training data?
- AThey show the expected or target output for each exampleCorrect answer
- BThey replace the need to split data into training and test sets
- CThey identify which rows are duplicates
- DThey specify the physical storage location of the data
Why this is the answer
Labels represent known outputs. Supervised learning uses labeled examples to learn a mapping from input features to the desired outcome.
- Question 25Multiple Select
Which two of the following scenarios are examples of supervised learning?
- ATraining a spam detector with emails previously marked as spam or not spamCorrect answer
- BSegmenting customers into groups based only on purchase behavior
- CPredicting house prices from historical sales with known pricesCorrect answer
- DReducing data to two dimensions for visualization without using labels
Why this is the answer
Spam detection and house price prediction use labeled examples, so they are supervised. Customer segmentation and dimensionality reduction are common unsupervised tasks.
- Question 26Interactive
The code snippet defines a small dataset with messages and categories. Based on this data, which two statements are correct?
- AThe dataset is appropriate for a supervised text classification taskCorrect answer
- BThe field 'category' is the label or outcomeCorrect answer
- CThe field 'text' stores the expected output
- DThe records contain numeric feature vectors already
Why this is the answer
Each record has text and a category label, making the data suitable for supervised classification. The category field is the target, while text is the input.
- Question 27Single Choice
What does the term inference mean in machine learning?
- AApplying a trained model to new data to produce predictionsCorrect answer
- BRemoving irrelevant records from the training set
- CAssigning labels manually to unlabeled training data
- DMeasuring CPU utilization during model deployment
Why this is the answer
Inference is the phase where a trained machine learning model receives new inputs and generates predictions or outputs.
OCI AI Portfolio
8 questions in this domain
- Question 28Single Choice
Which OCI AI service is designed to generate text using pretrained large language models in Oracle Cloud Infrastructure?
- AOCI Generative AICorrect answer
- BOCI Vision
- COCI Speech
- DOCI Language
Why this is the answer
OCI Generative AI provides hosted large language models for text generation. Vision, Speech, and Language are AI services for other types of input and analysis.
- Question 29Multiple Select
Select the two listed OCI offerings that are part of the OCI AI portfolio from the available choices.
- AOCI Generative AICorrect answer
- BOCI Autonomous Database
- COCI VisionCorrect answer
- DOCI Object Storage
Why this is the answer
OCI Generative AI and OCI Vision are services within the OCI AI portfolio. Autonomous Database and Object Storage are data and storage services, not AI services.
- Question 30Interactive
Review the endpoint in the code block. Which OCI AI service is represented by this request?
- AOCI LanguageCorrect answer
- BOCI Speech
- COCI Vision
- DOCI Generative AI
Why this is the answer
The code calls a dominant language detection method, which is an OCI Language natural language processing capability, not speech, vision, or generation.
- Question 31Single Choice
Which OCI AI service is designed to convert an uploaded audio file into written text?
- AOCI SpeechCorrect answer
- BOCI Vision
- COCI Language
- DOCI Anomaly Detection
Why this is the answer
OCI Speech provides automatic speech recognition and audio transcription. Vision processes images, Language analyzes text, and Anomaly Detection detects patterns in data.
- Question 32Multiple Select
Select the two capabilities that OCI Vision provides for image and document processing.
- AObject detection in imagesCorrect answer
- BOptical character recognition in documentsCorrect answer
- CSentiment analysis of text
- DSpeech-to-text transcription
Why this is the answer
OCI Vision includes object detection, image classification, and OCR. Sentiment analysis belongs to OCI Language, and speech-to-text belongs to OCI Speech.
- Question 33Interactive
Review the operation in the code block and choose which OCI AI service is used.
- AOCI SpeechCorrect answer
- BOCI Vision
- COCI Language
- DOCI Generative AI
Why this is the answer
The code block calls a transcription method for an audio file, which is the core function of the OCI Speech service.
- Question 34Single Choice
Which OCI AI service extracts text, key-value pairs, and tables from scanned documents and forms?
- AOCI Document UnderstandingCorrect answer
- BOCI Speech
- COCI Vision
- DOCI Anomaly Detection
Why this is the answer
OCI Document Understanding extracts text, tables, and key-value pairs from documents. Speech handles audio, Vision analyzes images, and Anomaly Detection examines time-series data.
- Question 35Multiple Select
From the options below, select the two true statements about the capabilities and deployment of OCI Generative AI.
- AIt offers hosted foundation models.Correct answer
- BIt can generate text and embeddings.Correct answer
- CIt requires on-premises installation.
- DIt only stores training images.
Why this is the answer
OCI Generative AI is a managed cloud service providing hosted foundation models for text generation and embeddings. It is not an on-premises product and is not for image storage.
OCI Generative AI Service and Oracle 23ai
5 questions in this domain
- Question 36Single Choice
Which OCI service provides a fully managed set of generative AI capabilities for building AI-powered applications?
- AOCI Generative AI ServiceCorrect answer
- BOCI Data Science
- COracle Analytics Cloud
- DOracle Integration Cloud
Why this is the answer
OCI Generative AI Service is the Oracle Cloud offering that provides managed access to pretrained models and infrastructure for building generative AI applications.
- Question 37Multiple Select
Which three capabilities are directly supported by Oracle 23ai AI Vector Search when working with embeddings and unstructured data?
- AStoring embeddings in a VECTOR columnCorrect answer
- BFinding semantically similar rows with SQLCorrect answer
- CSupporting RAG by retrieving relevant textCorrect answer
- DAutomatically training an LLM in the database
Why this is the answer
AI Vector Search supports storing vectors, querying by semantic similarity, and supplying retrieved content to RAG pipelines. It does not train large language models inside the database.
- Question 38Interactive
In Oracle 23ai, a developer wants to store text alongside its embedding. Review the SQL statement in the code block. Which action does it perform?
- AIt creates a table with a vector columnCorrect answer
- BIt updates existing rows with embeddings
- CIt drops the vector index for the table
- DIt queries vectors by semantic closeness
Why this is the answer
The CREATE TABLE statement creates a table named docs with a numeric key, a CLOB body, and an embedding vector.
- Question 39Single Choice
What type of OCI Generative AI Service resource provides isolated, tenancy-specific GPU compute for hosting and fine-tuning models?
- ADedicated AI clustersCorrect answer
- BOCI Object Storage buckets
- CCloud Guard responders
- DVCN gateways
Why this is the answer
Dedicated AI clusters in OCI Generative AI Service provide isolated GPU capacity for hosting models and for fine-tuning.
- Question 40Multiple Select
Which statements accurately describe retrieval-augmented generation (RAG) when used with OCI Generative AI Service and Oracle 23ai?
- ARetrieves relevant documents as context for the LLMCorrect answer
- BReduces retraining needs for new knowledgeCorrect answer
- CUses vector search to retrieve relevant contentCorrect answer
- DRequires the model to memorize documents during pretraining
Why this is the answer
RAG enriches prompts with relevant retrieved passages and can reduce the need for custom retraining. Oracle 23ai vector search supports this retrieval step.
Frequently asked questions
Everything about the Oracle Cloud Infrastructure 2026 AI Foundations Associate question bank and how to use it.