1. What is prompt engineering?
A) A software development methodology for building AI systems from scratch
B) The practice of designing and refining inputs to AI language models to
produce desired outputs
C) A hardware optimization technique for running large language models
faster
D) A method of training neural networks using reinforcement learning
✓ Correct Answer: B) The practice of designing and refining inputs to AI
language models to produce desired outputs
2. Which of the following best describes a 'prompt' in the context of large
language models (LLMs)?
A) The GPU memory allocated to a model during inference
B) A hyperparameter used during model training
C) The input text or instruction provided to a language model to guide its
response
D) The tokenization scheme used to encode model outputs
✓ Correct Answer: C) The input text or instruction provided to a
language model to guide its response
3. What does the term 'temperature' control in a language model's output?
A) The maximum number of tokens in the response
B) The randomness or creativity of the model's output
C) The speed at which the model generates tokens
D) The number of layers processed during inference
, ✓ Correct Answer: B) The randomness or creativity of the model's
output
4. A temperature setting of 0.0 in an LLM will typically produce:
A) Highly creative and unpredictable responses
B) Random, incoherent text
C) Deterministic, most-likely token outputs
D) Responses limited to a single sentence
✓ Correct Answer: C) Deterministic, most-likely token outputs
5. What is 'zero-shot prompting'?
A) Providing the model with many examples before asking a question
B) Asking the model to perform a task without providing any examples
C) Using a model that has not been fine-tuned on any data
D) Querying the model with an empty prompt
✓ Correct Answer: B) Asking the model to perform a task without
providing any examples
6. What is 'few-shot prompting'?
A) Providing several examples of input-output pairs in the prompt before
asking the model to complete a new task
B) Fine-tuning a model on a small dataset
C) Using a reduced model size to speed up inference
D) Limiting the model response to a few words
✓ Correct Answer: A) Providing several examples of input-output pairs
in the prompt before asking the model to complete a new task
7. Which prompting technique instructs the model to reason step-by-step
before providing a final answer?
A) Zero-shot prompting
B) Retrieval-augmented prompting
C) Chain-of-thought (CoT) prompting
D) Role prompting
✓ Correct Answer: C) Chain-of-thought (CoT) prompting
8. What is a 'system prompt' in the context of conversational AI?
A) The final message the user sends to end a conversation
B) A background instruction given to the model that sets its behavior, role, or
constraints before the conversation begins
, C) An automatically generated summary of a past conversation
D) A prompt that tests the model's system knowledge
✓ Correct Answer: B) A background instruction given to the model that
sets its behavior, role, or constraints before the conversation begins
9. What does 'hallucination' refer to in the context of LLMs?
A) The model generating responses faster than expected
B) The model producing confident but factually incorrect or fabricated
information
C) A visual artifact displayed in multimodal model outputs
D) The model refusing to answer a question
✓ Correct Answer: B) The model producing confident but factually
incorrect or fabricated information
10. Which of the following is the BEST strategy to reduce hallucinations in
LLM outputs?
A) Increase the temperature setting
B) Use shorter prompts
C) Provide relevant factual context in the prompt and ask the model to cite its
sources
D) Remove all examples from the prompt
✓ Correct Answer: C) Provide relevant factual context in the prompt and
ask the model to cite its sources
11. What is 'prompt injection'?
A) A technique to insert training data into a pre-trained model
B) A malicious attempt to override or hijack a model's instructions by
embedding adversarial text in the input
C) A method of compressing prompts to reduce token count
D) Automatically generating prompts using another AI model
✓ Correct Answer: B) A malicious attempt to override or hijack a model's
instructions by embedding adversarial text in the input
12. What is a 'token' in the context of language models?
A) A user authentication credential for API access
B) A unit of text (word, subword, or character) that the model processes
C) A reward signal used during reinforcement learning from human feedback
D) A single neuron in the transformer architecture
, ✓ Correct Answer: B) A unit of text (word, subword, or character) that
the model processes
13. What does 'context window' mean in LLMs?
A) The graphical user interface used to interact with the model
B) The maximum number of tokens the model can process in a single input-
output interaction
C) The number of layers in the transformer model
D) The portion of training data the model can recall
✓ Correct Answer: B) The maximum number of tokens the model can
process in a single input-output interaction
14. Which component of a well-structured prompt provides background
knowledge the model needs to answer correctly?
A) Instruction
B) Output format
C) Context
D) Temperature
✓ Correct Answer: C) Context
15. What is 'role prompting'?
A) Assigning specific tasks to different AI models in a pipeline
B) Instructing the model to adopt a specific persona or professional role to
shape its responses
C) A technique to suppress the model's default safety guidelines
D) Using different prompt templates for different user types
✓ Correct Answer: B) Instructing the model to adopt a specific persona
or professional role to shape its responses
16. What is 'retrieval-augmented generation' (RAG)?
A) Fine-tuning a model on retrieved internet data
B) A technique where the model retrieves relevant external documents and
uses them to augment its response
C) Generating prompts automatically from a database
D) A method of compressing the model by retrieving smaller parameter sets
✓ Correct Answer: B) A technique where the model retrieves relevant
external documents and uses them to augment its response
17. What does 'top-p' (nucleus) sampling control?
A) A software development methodology for building AI systems from scratch
B) The practice of designing and refining inputs to AI language models to
produce desired outputs
C) A hardware optimization technique for running large language models
faster
D) A method of training neural networks using reinforcement learning
✓ Correct Answer: B) The practice of designing and refining inputs to AI
language models to produce desired outputs
2. Which of the following best describes a 'prompt' in the context of large
language models (LLMs)?
A) The GPU memory allocated to a model during inference
B) A hyperparameter used during model training
C) The input text or instruction provided to a language model to guide its
response
D) The tokenization scheme used to encode model outputs
✓ Correct Answer: C) The input text or instruction provided to a
language model to guide its response
3. What does the term 'temperature' control in a language model's output?
A) The maximum number of tokens in the response
B) The randomness or creativity of the model's output
C) The speed at which the model generates tokens
D) The number of layers processed during inference
, ✓ Correct Answer: B) The randomness or creativity of the model's
output
4. A temperature setting of 0.0 in an LLM will typically produce:
A) Highly creative and unpredictable responses
B) Random, incoherent text
C) Deterministic, most-likely token outputs
D) Responses limited to a single sentence
✓ Correct Answer: C) Deterministic, most-likely token outputs
5. What is 'zero-shot prompting'?
A) Providing the model with many examples before asking a question
B) Asking the model to perform a task without providing any examples
C) Using a model that has not been fine-tuned on any data
D) Querying the model with an empty prompt
✓ Correct Answer: B) Asking the model to perform a task without
providing any examples
6. What is 'few-shot prompting'?
A) Providing several examples of input-output pairs in the prompt before
asking the model to complete a new task
B) Fine-tuning a model on a small dataset
C) Using a reduced model size to speed up inference
D) Limiting the model response to a few words
✓ Correct Answer: A) Providing several examples of input-output pairs
in the prompt before asking the model to complete a new task
7. Which prompting technique instructs the model to reason step-by-step
before providing a final answer?
A) Zero-shot prompting
B) Retrieval-augmented prompting
C) Chain-of-thought (CoT) prompting
D) Role prompting
✓ Correct Answer: C) Chain-of-thought (CoT) prompting
8. What is a 'system prompt' in the context of conversational AI?
A) The final message the user sends to end a conversation
B) A background instruction given to the model that sets its behavior, role, or
constraints before the conversation begins
, C) An automatically generated summary of a past conversation
D) A prompt that tests the model's system knowledge
✓ Correct Answer: B) A background instruction given to the model that
sets its behavior, role, or constraints before the conversation begins
9. What does 'hallucination' refer to in the context of LLMs?
A) The model generating responses faster than expected
B) The model producing confident but factually incorrect or fabricated
information
C) A visual artifact displayed in multimodal model outputs
D) The model refusing to answer a question
✓ Correct Answer: B) The model producing confident but factually
incorrect or fabricated information
10. Which of the following is the BEST strategy to reduce hallucinations in
LLM outputs?
A) Increase the temperature setting
B) Use shorter prompts
C) Provide relevant factual context in the prompt and ask the model to cite its
sources
D) Remove all examples from the prompt
✓ Correct Answer: C) Provide relevant factual context in the prompt and
ask the model to cite its sources
11. What is 'prompt injection'?
A) A technique to insert training data into a pre-trained model
B) A malicious attempt to override or hijack a model's instructions by
embedding adversarial text in the input
C) A method of compressing prompts to reduce token count
D) Automatically generating prompts using another AI model
✓ Correct Answer: B) A malicious attempt to override or hijack a model's
instructions by embedding adversarial text in the input
12. What is a 'token' in the context of language models?
A) A user authentication credential for API access
B) A unit of text (word, subword, or character) that the model processes
C) A reward signal used during reinforcement learning from human feedback
D) A single neuron in the transformer architecture
, ✓ Correct Answer: B) A unit of text (word, subword, or character) that
the model processes
13. What does 'context window' mean in LLMs?
A) The graphical user interface used to interact with the model
B) The maximum number of tokens the model can process in a single input-
output interaction
C) The number of layers in the transformer model
D) The portion of training data the model can recall
✓ Correct Answer: B) The maximum number of tokens the model can
process in a single input-output interaction
14. Which component of a well-structured prompt provides background
knowledge the model needs to answer correctly?
A) Instruction
B) Output format
C) Context
D) Temperature
✓ Correct Answer: C) Context
15. What is 'role prompting'?
A) Assigning specific tasks to different AI models in a pipeline
B) Instructing the model to adopt a specific persona or professional role to
shape its responses
C) A technique to suppress the model's default safety guidelines
D) Using different prompt templates for different user types
✓ Correct Answer: B) Instructing the model to adopt a specific persona
or professional role to shape its responses
16. What is 'retrieval-augmented generation' (RAG)?
A) Fine-tuning a model on retrieved internet data
B) A technique where the model retrieves relevant external documents and
uses them to augment its response
C) Generating prompts automatically from a database
D) A method of compressing the model by retrieving smaller parameter sets
✓ Correct Answer: B) A technique where the model retrieves relevant
external documents and uses them to augment its response
17. What does 'top-p' (nucleus) sampling control?
