Oracle Cloud Infrastructure 2026 Generative AI Professional

Last checked October 22, 2026 against the official Oracle exam page.

1. Question 1Oracle Cloud Infrastructure Generative AI Service

   Which OCI Generative AI option provides dedicated capacity for hosting a model with predictable performance for a production workload?

   1. AA local laptop with a GPU, which is used only for exploring prompts during development
   2. BOn-demand inference with no reserved capacity, which shares the service pool with other customers
   3. CA free trial account with no reserved capacity, which is intended only for short experiments
   4. DDedicated AI clusters, which provide reserved capacity for hosting models
   Show answer and explanationHide answer

   Correct answer: D

   Dedicated AI clusters provide reserved capacity for predictable performance in production. On-demand inference shares capacity, and trial or laptop environments are not production hosting.

2. Question 2Responsible AI and Governance

   Which practice helps reduce the risk that a generative AI application exposes personal data in its responses?

   1. AAdding all customer records to the prompt so that the model can answer questions about any customer
   2. BFiltering personal data from retrieved context and applying access controls to the data sources
   3. CStoring every user prompt and response in a public bucket so that the team can review them
   4. DDisabling access controls on the knowledge base so that the model can always find the data it needs
   Show answer and explanationHide answer

   Correct answer: B

   Filtering personal data and applying access controls limit exposure through responses. Adding all records, public storage and disabled access controls each increase exposure.

3. Question 3Prompt Engineering and Fine-Tuning Methodologies

   Which prompting technique supplies a few worked examples in the prompt so that the model follows the same format for a new input?

   1. AFew-shot prompting, which places examples of the desired input and output in the prompt
   2. BZero-shot prompting, which gives the instruction alone with no examples of the expected output
   3. CGradient clipping, which limits the size of weight updates during the training of the model
   4. DBatch inference, which processes many prompts together in a scheduled job overnight
   Show answer and explanationHide answer

   Correct answer: A

   Few-shot prompting uses examples in the prompt to guide format and style without retraining. Zero-shot has no examples, and clipping and batch inference are unrelated.

4. Question 4Retrieval-Augmented Generation

   In a retrieval-augmented generation system, what is the main role of the retriever?

   1. ATo find the document chunks most relevant to the user's query and supply them to the model
   2. BTo train the language model's weights on the company's private documents during each request
   3. CTo encrypt the user's query before it is sent to the language model for processing
   4. DTo store the final answer in a database so that it can be shown to the user later
   Show answer and explanationHide answer

   Correct answer: A

   The retriever finds relevant chunks for the query, which the model then uses to ground its answer. Training, encryption and answer storage are different functions.

5. Question 5Fundamentals of Large Language Models and Architecture

   Which component of a transformer-based LLM allows each token to weigh the relevance of other tokens in the same sequence?

   1. AThe learning rate schedule, which controls how quickly the model's weights change during training
   2. BThe tokenizer, which splits input text into token IDs before the model processes them
   3. CThe softmax output layer, which converts the final scores into probabilities over the vocabulary
   4. DThe self-attention mechanism, which computes weighted relationships between tokens in the sequence
   Show answer and explanationHide answer

   Correct answer: D

   Self-attention computes weighted relationships between tokens, which lets the model use context across the sequence. The tokenizer, output layer and learning rate schedule have different roles.