I85PREP
⌘ K
OCI GenAI Pro study materials
OCI GenAI Pro study resources

OCI GenAI Pro practice questions

One question at a time. Build your OCI GenAI Pro understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1Generative AI Fundamentals and OCI Generative AI Service

In OCI Generative AI, which concept describes the process of adapting a pretrained foundation model to a specific domain by updating its weights with your own labeled dataset?

  1. Semantic search
  2. Retrieval-augmented generation
  3. Fine-tuning Correct answer.
  4. Prompt engineering

Explanation

Fine-tuning continues training a pretrained model on domain-specific labeled data so its weights shift toward your task. Retrieval-augmented generation and prompt engineering change the input at inference time without altering model weights, and semantic search is a retrieval technique rather than a training method.

Question 2Generative AI Fundamentals and OCI Generative AI Service

Which OCI Generative AI feature lets an application ground model responses in documents stored in Object Storage without retraining the model?

  1. Custom model import
  2. Dedicated AI cluster scaling
  3. Retrieval-augmented generation Correct answer.
  4. Model fine-tuning

Explanation

Retrieval-augmented generation retrieves relevant passages from a knowledge base and supplies them as context so the model answers from your content. It requires no weight updates, unlike fine-tuning, and is unrelated to cluster scaling or importing model artifacts.

Question 3Generative AI Fundamentals and OCI Generative AI Service

A developer wants an OCI Generative AI playground session that keeps prior turns of a conversation so follow-up questions resolve references like "that policy". Which capability must be used?

  1. Embedding generation
  2. Chat with conversation history Correct answer.
  3. A larger max token setting
  4. A dedicated AI cluster

Explanation

Conversational context is preserved by sending prior turns as chat history so the model can resolve references across messages. Raising the token limit only allows longer output, a dedicated cluster affects hosting capacity, and embeddings produce vectors rather than dialogue state.

Question 4Generative AI Fundamentals and OCI Generative AI Service

Which statement accurately describes a foundation model in the context of OCI Generative AI?

  1. A model trained on broad data that can be adapted to many downstream tasks Correct answer.
  2. A load balancer that distributes inference requests across clusters
  3. A vector index that stores document embeddings for similarity search
  4. A rule-based classifier configured with hand-written decision trees

Explanation

Foundation models are pretrained on large, broad corpora and then adapted through prompting, RAG, or fine-tuning to many tasks. They are not hand-written rule systems, vector indexes, or traffic-distribution components.

Question 5Generative AI Fundamentals and OCI Generative AI Service

Which OCI Generative AI model family is designed to produce vector representations of text for semantic search and clustering?

  1. A chat model such as Cohere Command
  2. An embedding model such as Cohere Embed Correct answer.
  3. A text-generation model such as Meta Llama
  4. A reranking model such as Cohere Rerank

Explanation

Embedding models map text into dense vectors so similarity can be computed for search, clustering, and classification. Chat and text-generation models produce natural-language output, and rerank models reorder an existing candidate list rather than create embeddings.

Question 6Generative AI Fundamentals and OCI Generative AI Service

What is the primary purpose of a prompt in an OCI Generative AI inference request?

  1. To encrypt the request payload in transit
  2. To specify the instructions and input text the model conditions its response on Correct answer.
  3. To select which availability domain hosts the model
  4. To define the retention period for inference logs

Explanation

The prompt carries the instruction, context, and user input that the model conditions its generated output on. Encryption, placement, and log retention are handled by service and security configuration rather than by prompt content.

Question 7Generative AI Fundamentals and OCI Generative AI Service

Which parameter in an OCI Generative AI text generation request controls the maximum number of tokens the model may produce in its response?

  1. Temperature
  2. Max tokens Correct answer.
  3. Frequency penalty
  4. Top p

Explanation

Max tokens caps the length of the generated output. Temperature and top p shape randomness and sampling breadth, and frequency penalty discourages repetition; none of them bound output length.

Question 8Generative AI Fundamentals and OCI Generative AI Service

Which OCI Generative AI hosting option provides reserved, predictable inference capacity on hardware dedicated to a single tenancy?

  1. A public endpoint shared across customers
  2. The playground
  3. On-demand mode
  4. A dedicated AI cluster Correct answer.

Explanation

Dedicated AI clusters reserve compute hosted for your workloads, giving predictable performance and isolation. On-demand and shared public endpoints draw from pooled capacity, and the playground is an interactive console for experimentation rather than a capacity guarantee.

Question 9Generative AI Fundamentals and OCI Generative AI Service

A team must keep prompts and completions inside a private network path and avoid traversing the public internet. Which OCI Generative AI access method satisfies this requirement?

  1. The public inference endpoint
  2. The OCI console playground
  3. A pre-authenticated request on Object Storage
  4. A private endpoint in a VCN Correct answer.

Explanation

A private endpoint places the Generative AI service endpoint inside a VCN so traffic stays on the private network. Public endpoints and the console traverse public routes, and a pre-authenticated request grants temporary Object Storage access rather than private model inference.

Question 10Generative AI Fundamentals and OCI Generative AI Service

When building a RAG pipeline in OCI Generative AI, what is the role of the vector store?

  1. It queues batch inference jobs for later execution
  2. It caches authentication tokens for the Generative AI endpoint
  3. It stores the model weights used to serve inference requests
  4. It stores document embeddings so the most similar passages can be retrieved for a query Correct answer.

Explanation

A vector store indexes embeddings of document chunks and returns the nearest matches to a query embedding, which then become context for generation. Model weights live in the model artifact, token caching is an identity concern, and job queues are scheduling infrastructure.

Question 11Generative AI Fundamentals and OCI Generative AI Service

A RAG application returns answers that miss key facts because too few passages reach the model. Which adjustment most directly addresses this?

  1. Switch from an embedding model to a chat model
  2. Shorten the system prompt
  3. Decrease the temperature to zero
  4. Increase the number of retrieved chunks passed as context Correct answer.

Explanation

If relevant passages are not reaching the model, retrieving more chunks increases the chance the needed fact is present in context. Temperature affects randomness, swapping to a chat model removes the embedding capability, and shortening the system prompt does not add missing evidence.

Question 12Generative AI Fundamentals and OCI Generative AI Service

Which statement best distinguishes fine-tuning from retrieval-augmented generation on OCI Generative AI?

  1. Fine-tuning changes model weights, while RAG supplies external context at inference time Correct answer.
  2. Fine-tuning supplies external context, while RAG changes model weights
  3. Both methods avoid storing any data outside the model weights
  4. Both methods require retraining the embedding model for every query

Explanation

Fine-tuning performs additional training that alters weights, whereas RAG leaves weights untouched and injects retrieved context into each request. The reverse description is incorrect, and neither method implies retraining embeddings per query or forbids external data storage.

Question 13Generative AI Fundamentals and OCI Generative AI Service

A candidate is choosing between a base model and a fine-tuned custom model for a specialized classification task with thousands of labeled examples. Which consideration favors fine-tuning?

  1. The team needs the largest possible context window
  2. The task requires the model to internalize domain-specific label patterns Correct answer.
  3. The task requires the model to answer questions about documents that change hourly
  4. The team wants to avoid any training cost

Explanation

When abundant labeled data exists and the goal is to internalize task-specific patterns, fine-tuning adapts the weights to that behavior. Frequently changing documents favor RAG, avoiding training cost favors prompting, and context window size is a model capability unrelated to fine-tuning.

Question 14Generative AI Fundamentals and OCI Generative AI Service

Which OCI Generative AI component converts a user question into a vector so it can be compared against indexed document embeddings?

  1. The chat model
  2. The embedding model Correct answer.
  3. The rerank model
  4. The guardrails policy

Explanation

The embedding model encodes the query into the same vector space as the indexed chunks so similarity can be computed. Rerank reorders candidates, chat models generate text, and guardrails filter content rather than produce vectors.

Question 15Generative AI Fundamentals and OCI Generative AI Service

A developer sets temperature to a very low value for a summarization endpoint. What is the expected effect on responses?

  1. The model ignores the prompt instructions
  2. Output becomes more deterministic and focused Correct answer.
  3. Output becomes longer and more creative
  4. The context window expands automatically

Explanation

Low temperature narrows sampling toward the most probable tokens, producing consistent, focused output suited to summarization. It does not increase length or creativity, disable instruction following, or change the model's context window.

Question 16Generative AI Fundamentals and OCI Generative AI Service

Which OCI service is commonly paired with OCI Generative AI to store source documents that a RAG pipeline indexes?

  1. OCI Load Balancer
  2. OCI Object Storage Correct answer.
  3. OCI DNS
  4. OCI Bastion

Explanation

Object Storage holds the source documents and chunks that a RAG pipeline reads and embeds. Load Balancer, DNS, and Bastion provide traffic distribution, name resolution, and secure administrative access respectively, none of which store source content.

Question 17Generative AI Fundamentals and OCI Generative AI Service

A workload has steady, high-volume inference needs and strict latency targets. Which OCI Generative AI hosting choice is most appropriate?

  1. The console playground
  2. Manual retries against the public endpoint
  3. On-demand shared capacity
  4. A dedicated AI cluster sized for the expected load Correct answer.

Explanation

A dedicated AI cluster reserves capacity so steady high volume and latency targets are met without contending for shared resources. On-demand capacity is pooled, the playground is for experimentation, and retries do not solve capacity or latency problems.

Question 18Generative AI Fundamentals and OCI Generative AI Service

Which practice most improves the reliability of a structured-output prompt that must return JSON on OCI Generative AI?

  1. Provide an explicit schema and a short example in the prompt Correct answer.
  2. Raise the temperature to its maximum
  3. Remove all formatting instructions to save tokens
  4. Rely on the model to infer field names each time

Explanation

Stating the exact schema and including a compact example anchors the model to the required field names and structure. High temperature increases deviation, removing instructions reduces guidance, and letting field names vary undermines downstream parsing.

Question 19Generative AI Fundamentals and OCI Generative AI Service

What does the top p sampling parameter control in an OCI Generative AI generation request?

  1. The cumulative probability mass of tokens considered during sampling Correct answer.
  2. The maximum number of output tokens
  3. The billing rate per inference call
  4. The number of retrieved documents

Explanation

Top p restricts sampling to the smallest set of tokens whose cumulative probability reaches the threshold, controlling diversity. Output length is set by max tokens, retrieval count is a RAG setting, and billing is unrelated to sampling parameters.

Question 20Generative AI Fundamentals and OCI Generative AI Service

An application must answer questions over a knowledge base that is updated several times per day and must cite the current version. Which approach is most suitable?

  1. Fine-tune the model each time the knowledge base changes
  2. Rely on the model's pretrained knowledge alone
  3. Use RAG against a freshly indexed knowledge base Correct answer.
  4. Increase max tokens so the model recalls more

Explanation

RAG retrieves from an index that can be refreshed as documents change, so answers reflect the current version without retraining. Fine-tuning per update is slow and costly, larger output limits do not add current facts, and pretrained knowledge cannot reflect same-day changes.

Question 21Generative AI Fundamentals and OCI Generative AI Service

A RAG system retrieves many plausible passages but the final answer cites irrelevant ones. Which component should be adjusted to reorder candidates by relevance before generation?

  1. The max tokens setting
  2. The chat history buffer
  3. The dedicated AI cluster size
  4. The rerank model Correct answer.

Explanation

A rerank model scores retrieved candidates against the query and reorders them so the most relevant passages reach the generator. Chat history, cluster size, and output length do not change the relevance ordering of retrieved passages.

Question 22Generative AI Fundamentals and OCI Generative AI Service

A team wants to reduce hallucination in a summarization agent that must stay faithful to source documents. Which combination is most effective on OCI Generative AI?

  1. A high temperature plus a longer max tokens value
  2. Grounding with retrieved source text plus a low temperature Correct answer.
  3. Removing source documents and relying on the base model
  4. Increasing top p to one and disabling retrieval

Explanation

Supplying retrieved source text constrains the model to provided evidence, and low temperature reduces sampling-driven invention. High temperature, removing sources, and disabling retrieval all increase the chance of unsupported content.

Question 23Generative AI Fundamentals and OCI Generative AI Service

Why must the same embedding model be used for both indexing documents and embedding queries in an OCI Generative AI RAG pipeline?

  1. Vectors from different models occupy incompatible spaces, so similarity scores become meaningless Correct answer.
  2. Only one embedding model can be deployed per tenancy
  3. Reranking requires identical vector dimensions to function
  4. Query embeddings are always discarded before search

Explanation

Similarity search assumes query and document vectors share one coordinate space, which only holds when a single embedding model produces both. Tenancy deployment limits and query handling do not explain the requirement, and reranking operates on text pairs rather than requiring matched dimensions.

Question 24Generative AI Fundamentals and OCI Generative AI Service

A custom fine-tuned model must be served with guaranteed capacity separate from other tenants. Which OCI Generative AI configuration accomplishes this?

  1. Call the shared on-demand endpoint with a higher temperature
  2. Use the playground with a longer prompt
  3. Host the custom model on a dedicated AI cluster Correct answer.
  4. Store the model artifact in Object Storage only

Explanation

Custom models are hosted on dedicated AI clusters, which provide isolated, reserved capacity for serving. The shared endpoint does not guarantee isolation, storing an artifact alone does not serve it, and the playground is not a production hosting path.

Question 25Generative AI Fundamentals and OCI Generative AI Service

An architect must choose between prompt engineering, RAG, and fine-tuning for a task with no labeled data, a small stable instruction set, and a need to answer from a large private corpus. Which choice fits best?

  1. Fine-tuning on the private corpus
  2. Fine-tuning with no data
  3. RAG with well-crafted prompts Correct answer.
  4. Relying on default prompts without retrieval

Explanation

With no labeled data, fine-tuning is impractical, while RAG plus clear prompts grounds answers in the private corpus without training. Fine-tuning requires labeled examples, and default prompts without retrieval cannot access the private content.

Question 26Generative AI Fundamentals and OCI Generative AI Service

Which OCI Generative AI construct groups a base model together with its inference configuration so applications can call it through a stable identifier?

  1. A model catalog entry
  2. A dedicated AI cluster
  3. An endpoint Correct answer.
  4. A vector index

Explanation

In OCI Generative AI you create an endpoint that references a base or custom model and the serving configuration, then applications invoke that endpoint. The endpoint gives a stable, addressable target that can be updated without changing client code.

Question 27Generative AI Fundamentals and OCI Generative AI Service

Which OCI Generative AI capability lets an organization adapt a base model to its own labeled examples so the resulting model reflects domain-specific patterns?

  1. Dedicated AI clusters
  2. Fine-tuning Correct answer.
  3. Semantic search
  4. Prompt engineering

Explanation

Fine-tuning trains a base model further on curated labeled data, producing a custom model that captures task- or domain-specific behavior. Prompt engineering shapes behavior at inference time without changing weights, and the other options are hosting or retrieval features.

Question 28Generative AI Fundamentals and OCI Generative AI Service

Which OCI Generative AI model family is designed primarily to produce vector representations of text for similarity comparison?

  1. Chat models
  2. Reranking models
  3. Embedding models Correct answer.
  4. Guardrail models

Explanation

Embedding models convert text into dense vectors whose geometric proximity reflects semantic similarity, which is the basis for semantic search and retrieval. Chat models generate text, rerankers reorder candidate passages, and guardrails filter content.

Question 29Generative AI Fundamentals and OCI Generative AI Service

Which statement describes how the OCI Generative AI playground is used during development?

  1. It permanently deploys a custom model to production endpoints.
  2. It automatically fine-tunes a model from uploaded documents.
  3. It lets developers experiment interactively with prompts and parameters before coding. Correct answer.
  4. It replaces the need for an endpoint when serving production traffic.

Explanation

The playground is an interactive console for trying prompts, models, and parameters and observing outputs before committing to code. It does not provision production endpoints, train models automatically, or serve live application traffic.

Question 30Generative AI Fundamentals and OCI Generative AI Service

In an OCI Generative AI text generation request, which parameter controls the randomness of token selection during decoding?

  1. Frequency penalty
  2. Max tokens
  3. Stop sequences
  4. Temperature Correct answer.

Explanation

Temperature scales the sampling distribution, so lower values make output more deterministic and higher values make it more varied. Max tokens bounds length, frequency penalty discourages repetition, and stop sequences terminate generation.

Question 31Generative AI Fundamentals and OCI Generative AI Service

Which OCI Generative AI feature analyzes text and returns structured labels such as sentiment or key phrases without generating free-form prose?

  1. Text generation
  2. Text summarization
  3. Language analysis tasks such as classification Correct answer.
  4. Chat completion

Explanation

OCI Generative AI offers language analysis capabilities that return structured outputs like sentiment, classification labels, and key phrases. These differ from generative tasks that produce free-form or summarized prose.

Question 32Generative AI Fundamentals and OCI Generative AI Service

Which OCI Generative AI construct represents a private, isolated pool of compute used to host models with reserved capacity?

  1. A bastion host
  2. A load balancer
  3. A compartment
  4. A dedicated AI cluster Correct answer.

Explanation

A dedicated AI cluster allocates isolated compute resources so endpoints hosted on it do not share capacity with other tenants. Compartments organize resources, while bastions and load balancers are networking components.

Question 33Generative AI Fundamentals and OCI Generative AI Service

Which credential mechanism is the recommended way for an application running on OCI Compute to call OCI Generative AI without embedding long-lived keys?

  1. Instance principals Correct answer.
  2. Hard-coded user API keys in source code
  3. Shared console passwords
  4. Static auth tokens stored in environment files

Explanation

Instance principals let a compute instance obtain short-lived credentials automatically from the OCI identity service, removing the need to embed secrets. Hard-coded or static credentials create rotation and leakage risk.

Question 34Generative AI Fundamentals and OCI Generative AI Service

A developer wants the OCI Generative AI SDK to be invoked from a Python service. Which OCI identity concept must be configured so the SDK knows which tenancy and user context to use?

  1. A bucket lifecycle policy
  2. A cluster node pool
  3. A VCN route table
  4. An authentication configuration profile Correct answer.

Explanation

The SDK reads an authentication configuration profile that supplies tenancy, user, fingerprint, and key information (or instance principal settings). Route tables, bucket policies, and node pools are unrelated to SDK authentication.

Question 35Generative AI Fundamentals and OCI Generative AI Service

An application sends a prompt to a chat model and receives a response with a finish reason indicating the token limit was reached. What is the most direct cause?

  1. The temperature was set too low
  2. The prompt exceeded the context window
  3. The max tokens value was reached before the model finished Correct answer.
  4. The endpoint was hosted on a dedicated AI cluster

Explanation

A length-related finish reason typically means generation stopped because the configured max tokens limit was hit. Increasing the limit, within the model context window, allows longer completions.

Question 36Generative AI Fundamentals and OCI Generative AI Service

A team wants responses that are consistent and repeatable for a data-extraction task. Which parameter adjustment best supports that goal?

  1. Add many stop sequences
  2. Lower temperature toward zero Correct answer.
  3. Increase max tokens substantially
  4. Raise temperature and raise top p

Explanation

Lowering temperature narrows the sampling distribution, making the model favor its highest-probability tokens and produce more consistent outputs. Length and stop controls do not reduce sampling variability.

Question 37Generative AI Fundamentals and OCI Generative AI Service

An application must call OCI Generative AI from a region where the desired model is not offered. What is the practical implication?

  1. The request must target a region where that model is available Correct answer.
  2. Fine-tuning automatically replicates the model to every region
  3. The model can be invoked in any region regardless of availability
  4. Dedicated AI clusters make the model available in all regions

Explanation

Model availability varies by region, so requests must be routed to a region where the chosen model is offered. Hosting choices and fine-tuning do not override regional model availability.

Question 38Generative AI Fundamentals and OCI Generative AI Service

Which OCI Generative AI task is best suited for condensing a long support transcript into a short paragraph while preserving the main points?

  1. Embedding generation
  2. Text summarization Correct answer.
  3. Reranking
  4. Text classification

Explanation

Summarization condenses source text while retaining its key information, which matches the goal of shortening a transcript. Classification assigns labels, embeddings produce vectors, and reranking orders candidates.

Question 39Generative AI Fundamentals and OCI Generative AI Service

A developer needs the model to produce output strictly in a defined field order and format. Which approach is most reliable on OCI Generative AI?

  1. Remove all instructions to avoid constraining the model
  2. Specify the exact schema and an example in the prompt Correct answer.
  3. Rely on the model to infer the format from context
  4. Increase temperature so the model explores formats

Explanation

Providing an explicit schema plus a worked example anchors the model to the required structure. Raising temperature or omitting instructions increases variability and makes format compliance less likely.

Question 40Generative AI Fundamentals and OCI Generative AI Service

What distinguishes a custom model from a base model in OCI Generative AI?

  1. A custom model is always cheaper to invoke than a base model
  2. A custom model can only be used in the playground
  3. A custom model does not require an endpoint for inference
  4. A custom model results from fine-tuning a base model on customer data Correct answer.

Explanation

A custom model is produced by fine-tuning a supported base model with customer-provided training data, yielding weights adapted to that data. It still requires an endpoint to serve inference and does not guarantee lower cost.

Question 41Generative AI Fundamentals and OCI Generative AI Service

A team wants to experiment with a model's behavior across several parameter settings without writing code. Which OCI Generative AI resource supports this directly?

  1. The Generative AI playground Correct answer.
  2. Object Storage pre-authenticated requests
  3. A VCN security list
  4. A Functions application

Explanation

The playground provides an interactive interface for adjusting models and parameters and observing outputs immediately. The other options are storage, networking, and compute resources unrelated to model experimentation.

Question 42Generative AI Fundamentals and OCI Generative AI Service

Which statement correctly describes the relationship between an endpoint and a dedicated AI cluster in OCI Generative AI?

  1. A cluster is created automatically for every endpoint
  2. An endpoint can be hosted on a dedicated AI cluster for isolated capacity Correct answer.
  3. A dedicated AI cluster can host only one endpoint in total
  4. Endpoints and clusters are the same resource type

Explanation

An endpoint may be placed on a dedicated AI cluster so its inference runs on isolated, reserved compute. Clusters and endpoints are distinct resources, and clusters are not created implicitly for each endpoint.

Question 43Generative AI Fundamentals and OCI Generative AI Service

An application must send prompts to OCI Generative AI over the public internet while restricting who can call the service. Which control is most appropriate?

  1. IAM policies granting invoke permission to the specific group Correct answer.
  2. A higher top p value
  3. A longer system prompt
  4. A larger max tokens value

Explanation

IAM policies define which principals may invoke Generative AI resources, providing authorization control for callers. Generation parameters and prompt content affect output but not who is permitted to call the service.

Question 44Generative AI Fundamentals and OCI Generative AI Service

A developer is building a semantic search feature and must choose a model for converting both documents and queries into vectors. Which model type is appropriate?

  1. A chat model
  2. A summarization model
  3. A guardrail model
  4. An embedding model Correct answer.

Explanation

Embedding models map text into a vector space where semantic similarity corresponds to vector proximity, which is exactly what semantic search requires. Chat, summarization, and guardrail models serve different purposes.

Question 45Generative AI Fundamentals and OCI Generative AI Service

A team observes that outputs from a chat endpoint vary widely between identical requests even though the prompt is unchanged. Which explanation is most consistent with OCI Generative AI behavior?

  1. The endpoint is misconfigured and must be recreated
  2. The IAM policy is too permissive
  3. Sampling parameters allow stochastic token selection Correct answer.
  4. The model context window is too large

Explanation

Generative models sample from a probability distribution, so nonzero temperature or top p settings produce different outputs for identical prompts. Reducing randomness parameters makes responses more repeatable.

Question 46Generative AI Fundamentals and OCI Generative AI Service

Which statement best explains why a system prompt is useful in an OCI Generative AI chat application?

  1. It sets persistent instructions that shape every turn of the conversation Correct answer.
  2. It replaces the need for an endpoint
  3. It stores conversation history in Object Storage
  4. It permanently retrains the model weights

Explanation

A system prompt establishes role and behavioral guidance applied across the conversation, helping keep responses consistent. It does not alter model weights, replace endpoints, or persist data to storage.

Question 47Generative AI Fundamentals and OCI Generative AI Service

An architect must decide between on-demand inference and hosting on a dedicated AI cluster. Which factor most strongly favors the dedicated cluster?

  1. Occasional testing with unpredictable, low volume
  2. A desire to avoid creating any endpoint
  3. A need for isolated capacity with consistent performance for production traffic Correct answer.
  4. A requirement to use the playground more often

Explanation

Dedicated AI clusters reserve compute so production workloads get isolation and more predictable performance. On-demand inference suits sporadic or low-volume usage where reserved capacity is unnecessary.

Question 48Generative AI Fundamentals and OCI Generative AI Service

A team fine-tunes a model but continues to see weak results on a task that depends on facts changing daily. What is the best analysis?

  1. A larger max tokens value will refresh the model's knowledge
  2. Fine-tuning should be repeated hourly to stay current
  3. Raising temperature will update the model's facts
  4. Fine-tuning bakes in training-time knowledge and is poorly suited to rapidly changing facts Correct answer.

Explanation

Fine-tuning adjusts behavior and patterns from training data but does not provide a live feed of changing facts, so it is a poor fit for volatile information. Retrieval or prompt-supplied context is better for freshness.

Question 49Generative AI Fundamentals and OCI Generative AI Service

An application receives an error because the combined prompt and requested output exceed the model's context window. Which remediation is most appropriate?

  1. Reduce the prompt length or the requested max tokens Correct answer.
  2. Increase the temperature
  3. Switch to a lower top p value
  4. Add more stop sequences

Explanation

The context window bounds total tokens across input and output, so shortening the prompt or lowering the requested output length brings the request within limits. Sampling parameters do not change token capacity.

Question 50Generative AI Fundamentals and OCI Generative AI Service

A security reviewer asks how to limit which OCI Generative AI endpoints a development group may call. Which mechanism directly addresses this?

  1. Higher max tokens settings
  2. IAM policies scoped to the group and specific resources Correct answer.
  3. Larger dedicated AI clusters
  4. Shorter system prompts

Explanation

IAM policies bind principals to permitted actions on specified resources, so scoping policies to a group and specific endpoints restricts what that group can invoke. Generation settings and prompt content do not govern authorization.