Level 2–( AI ) Specialist | Artificial Intelligence |

Question 1 of 50

1. Question

1 point(s)

What does classification typically predict?

A category or class

Correct Incorrect Correct answer

A paragraph length

Correct Incorrect Correct answer

A storage location

Correct Incorrect Correct answer

A network address

Correct Incorrect Correct answer

Question 2 of 50

2. Question

1 point(s)

Why does tokenization matter for LLM applications?

It guarantees factual accuracy

Correct Incorrect Correct answer

It removes the need for prompts

Correct Incorrect Correct answer

It determines how text is represented and processed by the model

Correct Incorrect Correct answer

It prevents all hallucinations

Correct Incorrect Correct answer

Question 3 of 50

3. Question

1 point(s)

Why should deployed AI systems be monitored?

Models never change after deployment

Correct Incorrect Correct answer

Monitoring is only needed for hardware

Correct Incorrect Correct answer

Performance, data conditions, and risks can change over time

Correct Incorrect Correct answer

Testing becomes unnecessary after launch

Correct Incorrect Correct answer

Question 4 of 50

4. Question

1 point(s)

Which task is an example of regression?

Detecting spam or not spam

Correct Incorrect Correct answer

Predicting a property's numerical price

Correct Incorrect Correct answer

Grouping customers without labels

Correct Incorrect Correct answer

Generating a password

Correct Incorrect Correct answer

Question 5 of 50

5. Question

1 point(s)

What is an embedding?

A type of computer monitor

Correct Incorrect Correct answer

A password format

Correct Incorrect Correct answer

A physical AI device

Correct Incorrect Correct answer

A numerical representation capturing relationships or meaning in data

Correct Incorrect Correct answer

Question 6 of 50

6. Question

1 point(s)

What is explainability concerned with?

Increasing internet bandwidth

Correct Incorrect Correct answer

Reducing database size

Correct Incorrect Correct answer

Helping users understand relevant factors behind model outputs

Correct Incorrect Correct answer

Generating longer prompts

Correct Incorrect Correct answer

Question 7 of 50

7. Question

1 point(s)

What is clustering mainly used for?

Finding groups or patterns without predefined labels

Correct Incorrect Correct answer

Translating a document

Correct Incorrect Correct answer

Encrypting a database

Correct Incorrect Correct answer

Measuring processor speed

Correct Incorrect Correct answer

Question 8 of 50

8. Question

1 point(s)

What is the main purpose of vector search?

To remove duplicate keyboards

Correct Incorrect Correct answer

To replace model training completely

Correct Incorrect Correct answer

To find items with similar representations or meanings

Correct Incorrect Correct answer

To increase monitor resolution

Correct Incorrect Correct answer

Question 9 of 50

9. Question

1 point(s)

A bank uses an AI model for credit decisions. Which practice supports responsible deployment?

Hide the model's limitations

Correct Incorrect Correct answer

Remove all human governance

Correct Incorrect Correct answer

Use any available personal data

Correct Incorrect Correct answer

Monitor performance and assess fairness across relevant groups

Correct Incorrect Correct answer

Question 10 of 50

10. Question

1 point(s)

Why are training and test data usually separated?

To increase file size

Correct Incorrect Correct answer

To eliminate preprocessing

Correct Incorrect Correct answer

To evaluate performance on unseen data

Correct Incorrect Correct answer

To prevent the model from learning

Correct Incorrect Correct answer

Question 11 of 50

11. Question

1 point(s)

What does RAG stand for in many AI applications?

Random Automated Graphics

Correct Incorrect Correct answer

Retrieval-Augmented Generation

Correct Incorrect Correct answer

Real-time Application Gateway

Correct Incorrect Correct answer

Recursive Algorithmic Governance

Correct Incorrect Correct answer

Question 12 of 50

12. Question

1 point(s)

What is data leakage in Machine Learning?

When information that should not be available influences training or evaluation

Correct Incorrect Correct answer

When a hard drive physically breaks

Correct Incorrect Correct answer

When a prompt becomes shorter

Correct Incorrect Correct answer

When a model uses too few layers

Correct Incorrect Correct answer

Question 13 of 50

13. Question

1 point(s)

What is a neural network inspired by?

Database tables

Correct Incorrect Correct answer

Biological neural systems

Correct Incorrect Correct answer

Printer mechanisms

Correct Incorrect Correct answer

Internet protocols

Correct Incorrect Correct answer

Question 14 of 50

14. Question

1 point(s)

Why can RAG improve an LLM application?

It eliminates the need for source documents

Correct Incorrect Correct answer

It removes all security risks

Correct Incorrect Correct answer

It can provide relevant external information to support generation

Correct Incorrect Correct answer

It guarantees that the model can never hallucinate

Correct Incorrect Correct answer

Question 15 of 50

15. Question

1 point(s)

Why can class imbalance be a problem?

It always increases accuracy

Correct Incorrect Correct answer

It eliminates the need for labels

Correct Incorrect Correct answer

A model may perform poorly on underrepresented classes

Correct Incorrect Correct answer

It guarantees equal predictions

Correct Incorrect Correct answer

Question 16 of 50

16. Question

1 point(s)

What is the role of an activation function in a neural network?

To introduce nonlinear behavior into the model

Correct Incorrect Correct answer

To store user passwords

Correct Incorrect Correct answer

To compress every dataset

Correct Incorrect Correct answer

To replace the training process

Correct Incorrect Correct answer

Question 17 of 50

17. Question

1 point(s)

A company wants an LLM to answer questions from its internal policy documents. Which approach is particularly suitable?

RAG using authorized company documents

Correct Incorrect Correct answer

Disabling retrieval

Correct Incorrect Correct answer

Removing all documents

Correct Incorrect Correct answer

Training only on random public text

Correct Incorrect Correct answer

Question 18 of 50

18. Question

1 point(s)

What is transfer learning?

Moving a model from one computer to another manually

Correct Incorrect Correct answer

Reusing knowledge learned from one task or dataset to help another task

Correct Incorrect Correct answer

Transferring files between folders

Correct Incorrect Correct answer

Replacing all training data with random data

Correct Incorrect Correct answer

Question 19 of 50

19. Question

1 point(s)

What does an epoch represent during model training?

One complete pass through the training dataset

Correct Incorrect Correct answer

One model error

Correct Incorrect Correct answer

One input feature

Correct Incorrect Correct answer

One database table

Correct Incorrect Correct answer

Question 20 of 50

20. Question

1 point(s)

What is prompt engineering?

Manually labeling every image

Correct Incorrect Correct answer

Designing instructions and context to guide an AI model toward a desired output

Correct Incorrect Correct answer

Encrypting network traffic

Correct Incorrect Correct answer

Building computer processors

Correct Incorrect Correct answer

Question 21 of 50

21. Question

1 point(s)

What is fine-tuning?

Deleting the original model

Correct Incorrect Correct answer

Changing the computer screen

Correct Incorrect Correct answer

Further training a pretrained model on task-specific data

Correct Incorrect Correct answer

Searching the internet manually

Correct Incorrect Correct answer

Question 22 of 50

22. Question

1 point(s)

What is a learning rate?

A setting controlling how strongly model parameters are updated

Correct Incorrect Correct answer

The number of users in a system

Correct Incorrect Correct answer

The size of a hard drive

Correct Incorrect Correct answer

The number of output categories only

Correct Incorrect Correct answer

Question 23 of 50

23. Question

1 point(s)

Which prompt technique provides examples of the expected task format?

Random prompting

Correct Incorrect Correct answer

Few-shot prompting

Correct Incorrect Correct answer

Silent prompting

Correct Incorrect Correct answer

Hardware prompting

Correct Incorrect Correct answer

Question 24 of 50

24. Question

1 point(s)

When might fine-tuning be preferable to RAG?

When current external facts must always be retrieved

Correct Incorrect Correct answer

When no training examples exist

Correct Incorrect Correct answer

When the model should never change

Correct Incorrect Correct answer

When the desired behavior or task pattern needs to be learned consistently

Correct Incorrect Correct answer

Question 25 of 50

25. Question

1 point(s)

What is a hyperparameter?

A setting chosen before or around training that controls model behavior

Correct Incorrect Correct answer

A value always learned directly from every example

Correct Incorrect Correct answer

A type of training label

Correct Incorrect Correct answer

A hardware component

Correct Incorrect Correct answer

Question 26 of 50

26. Question

1 point(s)

What is zero-shot prompting?

Training a model with no data

Correct Incorrect Correct answer

Running two models together

Correct Incorrect Correct answer

Asking a model to perform a task without providing task-specific examples

Correct Incorrect Correct answer

Removing the model's parameters

Correct Incorrect Correct answer

Question 27 of 50

27. Question

1 point(s)

What is an AI evaluation benchmark?

A database password

Correct Incorrect Correct answer

A defined set of tests used to measure model performance

Correct Incorrect Correct answer

A type of neural network layer

Correct Incorrect Correct answer

A software installation package

Correct Incorrect Correct answer

Question 28 of 50

28. Question

1 point(s)

Why might a model use regularization?

To increase training errors intentionally

Correct Incorrect Correct answer

To reduce overfitting

Correct Incorrect Correct answer

To remove all training data

Correct Incorrect Correct answer

To eliminate the need for evaluation

Correct Incorrect Correct answer

Question 29 of 50

29. Question

1 point(s)

Why can structured output instructions be useful?

They guarantee factual accuracy

Correct Incorrect Correct answer

They eliminate model limitations

Correct Incorrect Correct answer

They make the expected response format clearer

Correct Incorrect Correct answer

They remove the need for validation

Correct Incorrect Correct answer

Question 30 of 50

30. Question

1 point(s)

Why should AI evaluation include realistic user scenarios?

They can reveal failures that simple laboratory tests may miss

Correct Incorrect Correct answer

They guarantee perfect deployment

Correct Incorrect Correct answer

They eliminate the need for monitoring

Correct Incorrect Correct answer

They prevent all unexpected behavior

Correct Incorrect Correct answer

Question 31 of 50

31. Question

1 point(s)

What does precision measure in a classification setting?

The proportion of predicted positives that are actually positive

Correct Incorrect Correct answer

The total number of training records

Correct Incorrect Correct answer

The number of model layers

Correct Incorrect Correct answer

The size of the dataset

Correct Incorrect Correct answer

Question 32 of 50

32. Question

1 point(s)

An AI application must return customer information in JSON format. What is the best practice?

Ask for “some information”

Correct Incorrect Correct answer

Clearly specify the required JSON structure and validate the result

Correct Incorrect Correct answer

Ignore formatting requirements

Correct Incorrect Correct answer

Allow any output without checking

Correct Incorrect Correct answer

Question 33 of 50

33. Question

1 point(s)

An LLM generates a convincing citation that does not exist. What should the developer prioritize?

Verification mechanisms and reliable source handling

Correct Incorrect Correct answer

Increasing the response length

Correct Incorrect Correct answer

Removing all source references

Correct Incorrect Correct answer

Trusting the model's confidence

Correct Incorrect Correct answer

Question 34 of 50

34. Question

1 point(s)

What does recall measure?

The proportion of actual positives that the model correctly identifies

Correct Incorrect Correct answer

The average prompt length

Correct Incorrect Correct answer

The number of model parameters

Correct Incorrect Correct answer

The amount of storage used

Correct Incorrect Correct answer

Question 35 of 50

35. Question

1 point(s)

What is an AI agent generally distinguished by?

Its use of printed documents only

Correct Incorrect Correct answer

Its dependence on manual typing for every step

Correct Incorrect Correct answer

Its inability to interact with systems

Correct Incorrect Correct answer

Its ability to pursue goals through actions, tools, or multiple steps

Correct Incorrect Correct answer

Question 36 of 50

36. Question

1 point(s)

What is prompt injection?

A method of compressing prompts

Correct Incorrect Correct answer

A technique for increasing monitor brightness

Correct Incorrect Correct answer

An attempt to manipulate an AI system through crafted input instructions

Correct Incorrect Correct answer

A type of neural network

Correct Incorrect Correct answer

Question 37 of 50

37. Question

1 point(s)

What is the main purpose of a validation dataset?

To store passwords

Correct Incorrect Correct answer

To tune and compare model choices during development

Correct Incorrect Correct answer

To replace training data

Correct Incorrect Correct answer

To generate hardware specifications

Correct Incorrect Correct answer

Question 38 of 50

38. Question

1 point(s)

A medical screening model must minimize missed positive cases. Which metric may be especially important?

Recall

Correct Incorrect Correct answer

File size

Correct Incorrect Correct answer

Token count

Correct Incorrect Correct answer

Latency only

Correct Incorrect Correct answer

Question 39 of 50

39. Question

1 point(s)

An AI agent is given access to an external database. What is an important security consideration?

Restrict permissions to only what the agent needs

Correct Incorrect Correct answer

Give unrestricted access to everything

Correct Incorrect Correct answer

Disable authentication

Correct Incorrect Correct answer

Store all credentials in prompts

Correct Incorrect Correct answer

Question 40 of 50

40. Question

1 point(s)

Which control can reduce risks from prompt injection in an AI application?

Give the model unrestricted permissions

Correct Incorrect Correct answer

Treat every user instruction as trusted

Correct Incorrect Correct answer

Separate trusted instructions from untrusted content and validate tool actions

Correct Incorrect Correct answer

Disable all monitoring

Correct Incorrect Correct answer

Question 41 of 50

41. Question

1 point(s)

A model performs very well on training data but poorly on unseen data. What is this most likely?

Underfitting

Correct Incorrect Correct answer

Normalization

Correct Incorrect Correct answer

Overfitting

Correct Incorrect Correct answer

Tokenization

Correct Incorrect Correct answer

Question 42 of 50

42. Question

1 point(s)

What is a confusion matrix used for?

To summarize classification predictions by actual and predicted classes

Correct Incorrect Correct answer

To create training prompts

Correct Incorrect Correct answer

To store neural network weights

Correct Incorrect Correct answer

To encrypt model outputs

Correct Incorrect Correct answer

Question 43 of 50

43. Question

1 point(s)

What is tool calling in an AI system?

Calling technical support by telephone

Correct Incorrect Correct answer

Replacing model training

Correct Incorrect Correct answer

Allowing a model to request an external function or service to perform a task

Correct Incorrect Correct answer

Creating a new processor

Correct Incorrect Correct answer

Question 44 of 50

44. Question

1 point(s)

A company wants to deploy an AI assistant connected to business systems. What is the most appropriate approach?

Connect every system immediately

Correct Incorrect Correct answer

Give the assistant administrator access

Correct Incorrect Correct answer

Deploy without evaluation to collect real-world data

Correct Incorrect Correct answer

Define the use case, limit permissions, test risks, monitor performance, and maintain human oversight

Correct Incorrect Correct answer

Question 45 of 50

45. Question

1 point(s)

What is a feature in a Machine Learning dataset?

A measurable input used by a model

Correct Incorrect Correct answer

A final certificate

Correct Incorrect Correct answer

A network cable

Correct Incorrect Correct answer

A model license

Correct Incorrect Correct answer

Question 46 of 50

46. Question

1 point(s)

What is a Large Language Model (LLM) primarily trained to do?

Model patterns in language and generate or predict text

Correct Incorrect Correct answer

Replace computer networks

Correct Incorrect Correct answer

Store documents without processing them

Correct Incorrect Correct answer

Control physical machinery directly

Correct Incorrect Correct answer

Question 47 of 50

47. Question

1 point(s)

Why is human approval useful in high-impact AI workflows?

It makes AI mathematically perfect

Correct Incorrect Correct answer

It provides oversight before consequential actions are taken

Correct Incorrect Correct answer

It eliminates the need for testing

Correct Incorrect Correct answer

It guarantees that data is unbiased

Correct Incorrect Correct answer

Question 48 of 50

48. Question

1 point(s)

Why is data preprocessing important?

It removes the need for testing

Correct Incorrect Correct answer

It guarantees perfect predictions

Correct Incorrect Correct answer

It makes all datasets identical

Correct Incorrect Correct answer

It prepares data for reliable model development

Correct Incorrect Correct answer

Question 49 of 50

49. Question

1 point(s)

What is a token in many language models?

A complete database

Correct Incorrect Correct answer

A unit of text processed by the model

Correct Incorrect Correct answer

A physical processor

Correct Incorrect Correct answer

A security certificate

Correct Incorrect Correct answer

Question 50 of 50

50. Question

1 point(s)

What is model drift?

A method for increasing storage

Correct Incorrect Correct answer

A change in real-world data or relationships that causes model performance to deteriorate

Correct Incorrect Correct answer

A model backup procedure

Correct Incorrect Correct answer

A type of prompt

Correct Incorrect Correct answer

Average score
Your score

Level 2–( AI ) Specialist | Artificial Intelligence

O.G. EDUCATION GROUP

ACADEMY OF OXFORD INTERNATIONAL ACADEMY OF GLASGOW FOR CULTURE & SCIENCES