MCQs

Disclaimer:

Educational Use Only:

The website links provided are to be used strictly for educational and knowledge-gaining purposes only.

No Misuse:

Do not misuse the links or the information contained therein for any commercial, illegal, or unethical activities.

Respect for Websites and Businesses:

Respect the intellectual property and business interests of the websites and their owners. Do not attempt to alter, copy, or redistribute content without proper authorization.

Acknowledge Value of Knowledge:

Recognize and value the knowledge and resources shared by these websites. Use the information responsibly and ethically.

No Endorsement:

The provision of these links does not imply endorsement of the linked sites, their content, or their products and services. Users are responsible for evaluating the suitability and accuracy of the information.

User Responsibility:

By accessing these links, you agree to use the information at your own risk and to comply with all applicable laws and website terms of use.

Overall MCQs for Reference

Which architecture is most commonly used as the backbone for LLMs?
a. Decision Trees
b. Convolutional Neural Networks
c. Transformers
d. Support Vector Machines
Answer: c. Transformersvitalflux+1
What does “pre-training” mean in the context of LLMs?
a. Training on specific customer data
b. Training a model on vast unlabeled data before fine-tuning
c. Inference after deploying a model
d. Hyperparameter optimization
Answer: b. Training a model on vast unlabeled data before fine-tuningyoutubedatacamp
The “attention mechanism” in transformers helps the LLM to:
a. Focus equally on all input tokens
b. Focus on the most relevant parts of input sequences
c. Eliminate dependencies
d. Increase training epochs
Answer: b. Focus on the most relevant parts of input sequencesdatacamp+1
Which is a primary use case for LLMs?
a. Numeric computation
b. Compiling code
c. Natural language generation
d. Image classification
Answer: c. Natural language generationnaviera101+1
Which technique addresses LLM bias and overfitting most directly?
a. Increasing model parameters
b. Data augmentation
c. Dropout regularization
d. Early stopping
Answer: b. Data augmentationmasteringllm.substack+1
One of the significant limitations of LLMs is:
a. Always requiring labeled data
b. Hallucination of facts
c. Linear scalability
d. Lack of parallelization
Answer: b. Hallucination of factstechnicalblog+1
Few-shot learning in LLMs refers to:
a. Learning from scratch
b. Learning from a few examples provided during prompt
c. Ensemble learning
d. Pre-training from millions of samples
Answer: b. Learning from a few examples provided during promptaiml+1
What is “fine-tuning” in LLMs?
a. Training from scratch
b. Adjusting hyperparameters before pre-training
c. Updating pre-trained weights on domain-specific data
d. Compressing the model
Answer: c. Updating pre-trained weights on domain-specific datayoutubedatacamp
What is the primary advantage of parameter sharing in transformers?
a. Lowering computation by sharing weights
b. Increasing compute demands
c. Limiting vocabulary
d. Creating separate models for each output
Answer: a. Lowering computation by sharing weightsvitalflux+1
Zero-shot learning with LLMs means:
a. Training on zero data points
b. Predicting on unseen tasks without specific training examples
c. Training with only one sample
d. Ignoring data augmentation
Answer: b. Predicting on unseen tasks without specific training examplesdatacamp+1
The “softmax” function in LLMs is mostly used for:
a. Backpropagation
b. Output normalization to probability distributions
c. Data augmentation
d. Loss minimization
Answer: b. Output normalization to probability distributionsvitalflux+1
Which is a commonly cited risk of deploying LLMs?
a. Reduced dependence on GPUs
b. Hallucinating plausible but inaccurate information
c. Redundancy in architecture
d. Strict rule-following
Answer: b. Hallucinating plausible but inaccurate informationtechnicalblog+1
LLMs are best suited for:
a. Tabular data regression
b. Time-series prediction
c. Natural language understanding and generation
d. Image segmentation
Answer: c. Natural language understanding and generationnaviera101+1
A key method to evaluate LLM performance is:
a. Confusion matrix
b. Perplexity
c. ROC curve
d. Dice score
Answer: b. Perplexitymasteringllm.substack+1
“Token” in LLM context typically means:
a. A model layer
b. A single neuron
c. A basic unit of text (word, subword, or character)
d. A complete sentence
Answer: c. A basic unit of text (word, subword, or character)datacamp+1
Which task is challenging for most LLMs?
a. Arithmetic reasoning over long contexts
b. Language translation
c. Text summarization
d. Generating poetry
Answer: a. Arithmetic reasoning over long contextssemanticscholar+1
What is Retrieval-Augmented Generation (RAG)?
a. Using LLMs to only retrieve documents
b. Augmenting model generation with external knowledge from retrieval
c. Encoding inputs as images
d. No use of context
Answer: b. Augmenting model generation with external knowledge from retrievalmasteringllm.substack+1
Which is NOT a component of the transformer architecture?
a. Self-Attention
b. Convolution Layer
c. Feedforward Layer
d. Positional Encoding
Answer: b. Convolution Layervitalflux+1
What is a “prompt” in the context of LLMs?
a. The input text or question given to guide model output
b. Model checkpoint
c. Output token list
d. Regularization term
Answer: a. The input text or question given to guide model outputtechnicalblog+1
Which is a major paradigm shift introduced by LLMs?
a. The use of tabular learning
b. Sequence-to-sequence learning
c. The use of transformer architecture for text
d. Greedy decoding
Answer: c. The use of transformer architecture for textdatacamp+1
During inference, the temperature parameter controls:
a. Training rate
b. Output randomness and creativity
c. Token alignment
d. Hardware demands
Answer: b. Output randomness and creativitymasteringllm.substack+1
Which dataset is often used to train LLMs?
a. MNIST
b. CIFAR-10
c. Common Crawl
d. ImageNet
Answer: c. Common Crawlvitalflux+1
The main benefit of pre-trained embeddings in LLMs is:
a. Visualizing data
b. Initializing weights with general language knowledge
c. Randomized input processing
d. Decreasing learning rates
Answer: b. Initializing weights with general language knowledgedatacamp+1
Why is “layer normalization” used in transformers?
a. To decrease layer count
b. To stabilize and speed up training
c. To randomize outputs
d. To increase bias
Answer: b. To stabilize and speed up trainingvitalflux+1
Which of the following best describes hallucination in LLMs?
a. Outputting real numbers only
b. Generating plausible but incorrect or unverifiable information
c. Overfitting to training data
d. Pruning weights
Answer: b. Generating plausible but incorrect or unverifiable informationtechnicalblog+1
“Prompt engineering” is the practice of:
a. Adjusting loss functions for convergence
b. Designing optimal inputs to get desired model outputs
c. Memory optimization
d. Parameter tuning
Answer: b. Designing optimal inputs to get desired model outputstechnicalblog+1
Which strategy can reduce LLM inference cost?
a. Increasing model depth
b. Quantization and model pruning
c. Increasing learning rate
d. Using larger vocabularies
Answer: b. Quantization and model pruningmasteringllm.substack+1
Which loss function is commonly used in LLMs for language modeling?
a. Mean Squared Error
b. Cross-Entropy Loss
c. Hinge Loss
d. Dice Loss
Answer: b. Cross-Entropy Losstestbook+1
LLM “tokenization” aims to:
a. Split text into manageable pieces for the model
b. Merge all words
c. Assign data labels
d. Create confusion matrix
Answer: a. Split text into manageable pieces for the modeldatacamp+1
RLHF stands for:
a. Reinforcement Learning from Human Feedback
b. Random Length Hyperparameter Factor
c. Recursive Layer Hidden Function
d. Read Level Hash Filter
Answer: a. Reinforcement Learning from Human Feedbackaiml+1
How is “context length” relevant in LLMs?
a. It is irrelevant
b. Determines how much prior input model can consider
c. Controls batch size only
d. Fixes model learning rate
Answer: b. Determines how much prior input model can considermasteringllm.substack+1
A benefit of model quantization for LLMs is:
a. Larger disk storage
b. Reduced memory and faster inference
c. Increased number of parameters
d. No change
Answer: b. Reduced memory and faster inferencemasteringllm.substack+1
“Positional encodings” help transformers to:
a. Ignore position of tokens
b. Write code
c. Capture sequence order of input data
d. Run faster
Answer: c. Capture sequence order of input datavitalflux+1
Which optimization algorithm is especially common for LLM training?
a. Stochastic Gradient Descent (SGD)
b. Adam
c. RMSprop
d. Adagrad
Answer: b. Adamtestbook+1
What does “autogressive” mean in LLMs?
a. Outputting all tokens at once
b. Generating sequence tokens one at a time using previous output as context
c. Only training on images
d. Not using attention
Answer: b. Generating sequence tokens one at a time using previous output as contextdatacamp+1
Which safety mechanism helps moderate LLM output?
a. Data normalization
b. Output filtering for toxic or unsafe text
c. Dropout
d. Increasing context length
Answer: b. Output filtering for toxic or unsafe texttechnicalblog+1
The main purpose of “embedding layers” is to:
a. Create visualizations
b. Represent words or tokens as dense vectors
c. Tokenize input
d. Compute loss
Answer: b. Represent words or tokens as dense vectorsvitalflux+1
What is the likely impact of increasing parameter count in LLMs?
a. Lower language understanding
b. Better generalization and capability (and hardware demands)
c. Inability to use attention
d. Underfitting
Answer: b. Better generalization and capability (and hardware demands)datacamp+1
When do LLMs use “beam search”?
a. During training
b. During inference for text generation to optimize next word selection
c. For model quantization
d. For dropout
Answer: b. During inference for text generation to optimize next word selectionvitalflux+1
The main role of the decoder in a transformer (when present) is to:
a. Encode input
b. Generate output sequence from encoded input
c. Prune weights
d. Quantize layers
Answer: b. Generate output sequence from encoded inputdatacamp+1
“Multi-head attention” in transformers enables:
a. Training fewer layers
b. Capturing different types of relationships in parallel
c. Overfitting intentionally
d. Data shuffling
Answer: b. Capturing different types of relationships in parallelvitalflux+1
Which is a bottleneck for scaling up LLMs?
a. Data scarcity
b. Hardware and computational resources
c. Overfitting
d. Limited hyperparameters
Answer: b. Hardware and computational resourcesdatacamp+1
The “hidden state” in a transformer refers to:
a. A model token
b. Internal representation of data between layers
c. GPU allocation
d. Output label
Answer: b. Internal representation of data between layersvitalflux+1
What is the effect of “dropout” during LLM training?
a. Reduces overfitting by randomly disabling units
b. Increases data bias
c. Adds more layers
d. Prunes datasets
Answer: a. Reduces overfitting by randomly disabling unitstestbook+1
What is a primary goal of “chain-of-thought” prompting?
a. Increase hardware costs
b. Guide LLMs to reason step-by-step for better answers
c. Decrease vocabulary
d. Lower accuracy
Answer: b. Guide LLMs to reason step-by-step for better answersmasteringllm.substack+1
One LLM problem area is:
a. Always using supervised learning
b. Vulnerability to adversarial prompting
c. Parallel computation only
d. Output tokens are always correct
Answer: b. Vulnerability to adversarial promptingopenreview+1
What is “embedding similarity search” in LLM-based systems?
a. Searching for different model architectures
b. Finding documents with similar semantic meaning via vector comparisons
c. Computing loss values
d. Increasing model batch size
Answer: b. Finding documents with similar semantic meaning via vector comparisonsmasteringllm.substack+1
What does “context window” mean in LLMs?
a. Output display size
b. The maximum number of tokens the model processes in one sequence
c. Optimization loop
d. Validation loss
Answer: b. The maximum number of tokens the model processes in one sequencemasteringllm.substack+1
Why are LLMs important for AI research?
a. For binary classification tasks only
b. They show strong generalization for natural language understanding and generation
c. They only work on structured data
d. No application beyond chatbots
Answer: b. They show strong generalization for natural language understanding and generationdatacamp+1
What is a role of “explainability” in LLMs?
a. Making output more random
b. Helping users and developers trust and understand decisions
c. Increasing hallucinations
d. Decreasing inference speed
Answer: b. Helping users and developers trust and understand decisionsvitalflux+1

Blog

Systemethic Technologies

Leave a Reply Cancel reply

Useful links

Other Terms & Policies

Contact Info