# Free 1Z0-1122-26 practice questions

> 10 free 1Z0-1122-26 practice questions with the correct answer and an explanation.

Page: https://devopscodelab.com/exams/oracle-1z0-1122-26/free-practice-questions/
Practice exams: https://devopscodelab.com/exams/oracle-1z0-1122-26/

### Question 1 of 10

A retail manager asks how artificial intelligence, machine learning, and deep learning relate to each other. Which statement describes the relationship correctly?

- A. Deep learning and machine learning are independent fields that only overlap in computer vision
- B. Machine learning is a subset of deep learning, and both sit alongside AI as separate disciplines
- C. Deep learning is a subset of machine learning, which is itself a subset of artificial intelligence
- D. Artificial intelligence is a subset of machine learning that adds rule-based logic on top of models

**Answer:** C

**Explanation:** AI is the broad goal of machines performing tasks that need human-like intelligence, ML is the subset that learns from data, and DL is the subset of ML that uses multi-layer neural networks. Machine learning is not inside deep learning; the nesting runs the other way. AI is the umbrella, not a subset of ML, and deep learning is a form of machine learning rather than an independent field limited to vision.

### Question 2 of 10

What is the main idea behind ensemble methods such as random forests?

- A. Training one very deep tree until it fits every example
- B. Labeling unlabeled data with a single pretrained model
- C. Combining many models so errors cancel out
- D. Removing features until only the strongest one remains

**Answer:** C

**Explanation:** Ensembles combine predictions from many models, such as many trees in a random forest, which reduces variance and usually improves accuracy. One fully grown tree tends to overfit, keeping one feature discards useful signal, and pseudo-labeling with a single model is a semi-supervised trick rather than an ensemble.

### Question 3 of 10

During training, a model's training loss keeps falling while its validation loss starts rising after epoch 12. What is the BEST response?

- A. Remove the validation set and rely on training loss as the main signal
- B. Use early stopping around the epoch where validation loss was lowest
- C. Keep training for more epochs until the validation loss also falls
- D. Increase the learning rate so the model escapes the rising validation loss

**Answer:** B

**Explanation:** Diverging training and validation loss means the model has started to overfit, so stopping near the validation minimum keeps the best-generalizing model. More epochs deepen overfitting, a higher learning rate does not address overfitting, and dropping validation removes the only signal that revealed the problem.

### Question 4 of 10

In a very deep network that uses sigmoid activations, the early layers barely learn because their gradients become tiny. What is this problem called?

- A. The exploding learning rate problem
- B. The class imbalance problem
- C. The cold start problem
- D. The vanishing gradient problem

**Answer:** D

**Explanation:** Repeated multiplication of small sigmoid derivatives shrinks gradients toward zero in early layers, which is the vanishing gradient problem; ReLU, residual connections, and normalization help. The learning rate is a set value rather than something that explodes, cold start concerns recommenders with new users, and class imbalance concerns label distribution.

### Question 5 of 10

Which loss function is commonly used when training a neural network for multiclass classification?

- A. Hinge loss on regression
- B. Cross-entropy loss
- C. Mean absolute error
- D. Mean squared error

**Answer:** B

**Explanation:** Cross-entropy compares predicted class probabilities with the true class and is standard for classification with softmax outputs. Mean squared and mean absolute error are typical for regression, and hinge loss is associated with margin classifiers such as SVMs rather than regression.

### Question 6 of 10

A team wants an LLM to always reply in a specific brand voice and output format across thousands of requests, and prompt instructions alone give inconsistent results. Which approach is MOST appropriate?

- A. Use RAG to retrieve brand guidelines and append them to every prompt
- B. Switch to zero-shot prompts without any examples to reduce prompt length
- C. Increase the top-p value so the model samples from more candidate tokens
- D. Fine-tune the model on curated examples of the desired style and format

**Answer:** D

**Explanation:** When behavior such as tone and structure must be consistent and prompting falls short, fine-tuning on curated examples teaches the style directly. RAG is best for injecting knowledge rather than reliably changing style, higher top-p adds variety rather than consistency, and removing examples usually makes formatting less consistent.

### Question 7 of 10 (choose 2)

A legal firm's LLM summaries sometimes cite clauses that do not appear in the contract being summarized. Which TWO changes are MOST likely to reduce these invented clauses? (Choose two.)

- A. Lower the temperature so the model favors its most probable, conservative wording
- B. Shorten the contract excerpt to its first page so the prompt uses fewer tokens
- C. Raise top-p toward 1.0 so the model samples from a wider range of tokens
- D. Supply the contract text in the prompt and instruct the model to use only that text
- E. Remove the system prompt so the model writes without any formatting rules

**Answer:** A, D

**Explanation:** Grounding the model in the actual contract with an instruction to stay within it, and lowering temperature to reduce creative sampling, both cut down invented content. A wider top-p increases variety, removing the system prompt drops useful constraints, and truncating the contract removes the very clauses the summary needs.

### Question 8 of 10

A business analyst using Oracle Analytics Cloud wants to apply ML predictions to a data set without writing code. Which approach fits BEST?

- A. Use OCI Speech to transcribe the data set into spoken audio
- B. Ask a DBA to create HNSW vector indexes on the source tables
- C. Use the built-in ML features in Oracle Analytics Cloud
- D. Deploy an LLM on a bare metal GPU and write a Python client

**Answer:** C

**Explanation:** Oracle Analytics includes built-in ML features, such as training and applying models in data flows, aimed at analysts. Vector indexes support similarity search rather than predictions, a self-hosted LLM requires coding and infrastructure, and Speech converts audio and text rather than predicting from data sets.

### Question 9 of 10

Which statement about customer data in OCI Generative AI is accurate?

- A. Every prompt is added to the public base model's training data to improve it for all customers
- B. Prompts and fine-tuning data are not shared with model providers to train base models
- C. Customer data must be uploaded to the model provider's own cloud before fine-tuning
- D. Fine-tuned custom models are shared with other tenancies in the same region by default

**Answer:** B

**Explanation:** OCI Generative AI is designed so customer prompts, responses, and training data stay within the customer's control and are not used by model providers to train base models; custom models stay private to the tenancy. Sharing prompts publicly, sharing custom models across tenancies, or uploading to the provider's cloud would contradict that design.

### Question 10 of 10

Which OCI service converts written text into natural-sounding audio, for example to read announcements aloud in an app?

- A. OCI Vision text detection
- B. OCI Speech text-to-speech
- C. OCI Digital Assistant skills
- D. OCI Language translation

**Answer:** B

**Explanation:** OCI Speech offers text-to-speech in addition to speech-to-text transcription. Language translation converts between languages in text, Vision text detection reads text from images, and Digital Assistant skills handle conversations rather than synthesizing audio.

Ready for a timed run? Take the free 15-question 1Z0-1122-26 mini exam: https://devopscodelab.com/app/exams/oracle-1z0-1122-26/mini/

DevOpsCode Lab is not affiliated with Oracle. 1Z0-1122-26 and related names are trademarks of their owners. We write every question ourselves; these are not real exam questions.
