Oracle 1Z0-1158-26 Certification Sample Questions and Answers

1Z0-1158-26 PDF, 1Z0-1158-26 Dumps PDF Free Download, 1Z0-1158-26 Latest Dumps Free PDF, Cloud Infrastructure Enterprise AI Professional PDF DumpsThe Oracle Cloud Infrastructure Enterprise AI Professional (1Z0-1158-26) Sample Question Set is designed to help you prepare for the Oracle Cloud Infrastructure Certified Enterprise AI Professional certification exam. To become familiar with the actual Oracle Certification exam environment, we suggest you try our Sample Oracle 1Z0-1158-26 Certification Practice Exam.

This Oracle Cloud Infrastructure Enterprise AI Professional certification sample practice test and sample question set are designed for evaluation purposes only. If you want to test your Oracle 1Z0-1158-26 knowledge to identify your areas of improvement and get familiar with the actual exam format, we suggest you prepare with the Premium Oracle Cloud Infrastructure Certified Enterprise AI Professional Certification Practice Exam. Our team of Oracle Cloud Infrastructure experts has designed Questions-Answers for this premium practice exam by collecting inputs from recently certified candidates. Our premium Oracle 1Z0-1158-26 certification practice exam will boost your confidence as well as your actual Oracle Cloud Infrastructure Enterprise AI Professional exam result.

Oracle 1Z0-1158-26 Sample Questions:

01. An insurer uses a chat model to decide whether claim line items are covered under a policy excerpt included in the prompt. Reviewers report two problems: the model often returns a free-text paragraph instead of the required COVERED or NOT COVERED verdict with a cited clause, and on clauses with several conditions it skips some conditions and reaches wrong verdicts.
Which two prompt refinements address these problems?
(Choose two.)
a) Remove the policy excerpt so the prompt is shorter and more focused
b) Instruct the model to check each condition of the clause before the verdict
c) Include worked examples that show the exact verdict and citation format
d) Add a presence penalty so the model avoids reusing words from the clause
e) Raise the temperature so the model explores more interpretations
 
02. A bank plans to fine-tune a custom model on internal documents. Its security team requires that the bank control the encryption keys protecting the training data, in addition to the provider's default encryption.
Which option does OCI Generative AI document for this requirement?
a) Customer-managed keys in OCI Vault, on top of Oracle-managed AES-256 encryption
b) Oracle-managed AES-256 encryption only, since the service accepts no other keys
c) TLS 1.2 on every request, which places the stored data under the bank's control
d) Training data kept on the model endpoint and encrypted with the endpoint API key
 
03. An order-support agent receives "Where is order 55120?". It calls an order-lookup tool, and the tool returns {"status": "not_found"}. The agent follows the perceive-reason-act-observe cycle.
What should the agent do next?
a) Return the tool output to the user as the final answer, without any further reasoning
b) Add the result to its context and reason about the next step to take
c) Repeat the same lookup call until the tool returns a status other than not found
d) Discard the tool result and restart the cycle from the original user message
 
04. A research agent is asked to compile competitor pricing. Logs show it repeatedly calls a web-search tool with slightly reworded queries, never produces an answer, and eventually hits the request timeout. Each search returns relevant results.
Which change most directly prevents this behavior?
a) Raise the model temperature so each iteration produces more varied queries
b) Increase the request timeout so the agent has more time to finish the loop
c) Add more search tools, such as a news search and a pricing database
d) Add a stopping condition, such as a goal check and an iteration cap
 
05. An application calls a pretrained chat model in on-demand mode. Before each prompt reaches the model, the team wants to detect prompt-injection attempts and personally identifiable information, and let the application decide whether to block, redact or continue.
How should the team apply guardrails?
a) Configure Block mode on the model's endpoint and let OCI reject flagged prompts
b) Rely on the output filtering that is built into the foundational models
c) Call the ApplyGuardrails API on each prompt and act on the results
d) Enable PII and injection guardrails in the project settings
 
06. A manufacturer must translate equipment manuals from English into German. The model has to take in each complete source paragraph before producing the translated paragraph, and the output is a new sequence, of a different length, in another language.
Which LLM architecture is designed for exactly this kind of task?
a) Encoder-only, which builds a bidirectional representation of the whole input
b) Decoder-only, which reads the prompt, then generates one token at a time
c) Encoder-only with a classification head, which labels each paragraph
d) Encoder-decoder, which encodes the input, then decodes a new sequence
 
07. An agent's request declares an mcp tool that points at a partner's remote MCP server. The call fails. The partner confirms that the server is reachable from the internet, that the token is valid, and that the server exposes its tools over the SSE transport only.
What is the cause?
a) MCP Calling reaches servers inside the caller's tenancy, and this one is external
b) MCP Calling supports only Streamable HTTP, not the SSE transport
c) MCP Calling requires each server tool to be declared as a function tool
d) MCP Calling requires the server to be registered in the project settings
 
08. A hosted agent's container runs correctly in local testing, where it serves requests on 127.0.0.1:5000. After deployment it never receives traffic, and the platform reports that the container is not ready.
Which change to the container is required?
a) Listen on 127.0.0.1 port 443 and terminate TLS inside the container
b) Set the PORT environment variable to 5000 in the application
c) Listen on 0.0.0.0 port 8080 and return 200 from the /ready endpoint
d) Add a volume for startup logs so the platform can confirm readiness
 
09. An administrator adds a policy that lets the analysts group use generative-ai-chat only where target.model.id matches two approved models. Testing shows that analysts can still call every chat model. An older policy in the same compartment reads allow group analysts to use generative-ai-family in compartment ai-lab.
What should the administrator do?
a) Add an ALL condition that excludes the other models as an explicit deny
b) Remove or narrow the older family grant, because IAM policies are cumulative
c) Change the new policy's verb from use to manage so that it takes precedence
d) Move the conditional policy above the older one, because statements are evaluated in order
 
10. A developer configures the OpenAI SDK with the OCI Generative AI /openai/v1 base URL for the right region and an OCI Generative AI API key that has the required permission policy. The client sets only the base URL and the API key. Calls to the Responses API fail before any model runs.
What is missing?
a) A project OCID, as OpenAI-compatible calls in the service need a project
b) An IAM API signing key pair, because Generative AI API keys are not accepted
c) A vector store ID, because each Responses API request must declare a tool
d) A dedicated AI cluster, because the Responses API serves only dedicated models

Answers:

Question: 01

Answer: b, c

Question: 02

Answer: a

Question: 03

Answer: b

Question: 04

Answer: d

Question: 05

Answer: c

Question: 06

Answer: d

Question: 07

Answer: b

Question: 08

Answer: c

Question: 09

Answer: b

Question: 10

Answer: a

 

No votes yet