example.json

[
    {
        "question": "Which of the following best describes a Large Language Model (LLM)?",
        "answer_a": "A model specializing in language recognition",
        "answer_b": "A massive neural network that understands and generates human language",
        "answer_c": "A model exclusively used for language data tasks like summarization or classification",
        "answer_d": "A rule-based chatbot used for conversations",
        "correct_answer": "B"
    },
    {
        "question": "LLMs are typically:",
        "answer_a": "Pre-trained on small, curated datasets",
        "answer_b": "Trained on large text corpora to capture linguistic patterns",
        "answer_c": "Trained purely on translation tasks",
        "answer_d": "Designed to function solely with GPU resources",
        "correct_answer": "B"
    },
    {
        "question": "Which of the following is a common architecture for LLMs?",
        "answer_a": "Convolutional Neural Networks (CNNs)",
        "answer_b": "Transformer",
        "answer_c": "Recurrent Neural Networks (RNNs) with LSTM",
        "answer_d": "Support Vector Machines",
        "correct_answer": "B"
    },
    {
        "question": "What does it mean when we say LLMs are \"autoregressive\"?",
        "answer_a": "They regress to the mean to reduce variance",
        "answer_b": "They generate text by predicting the next token based on previous tokens",
        "answer_c": "They can only handle labeled data",
        "answer_d": "They can output text only after the entire input is known at once",
        "correct_answer": "B"
    },
    {
        "question": "Which of these is NOT a common use of LLMs?",
        "answer_a": "Summarizing content",
        "answer_b": "Generating code",
        "answer_c": "Playing strategy games like chess or Go",
        "answer_d": "Conversational AI",
        "correct_answer": "C"
    },
    {
        "question": "Which of the following best describes a \"special token\"?",
        "answer_a": "A token that makes the model forget all context",
        "answer_b": "A model signature required for API calls",
        "answer_c": "A token that helps segment or structure the conversation in the model",
        "answer_d": "A token that always represents the end of text",
        "correct_answer": "C"
    },
    {
        "question": "What is the primary goal of a \"chat template\"?",
        "answer_a": "To force the model into a single-turn conversation",
        "answer_b": "To structure interactions and define roles in a conversation",
        "answer_c": "To replace the need for system messages",
        "answer_d": "To store prompts into the model's weights permanently",
        "correct_answer": "B"
    },
    {
        "question": "How do tokenizers handle text for modern NLP models?",
        "answer_a": "By splitting text into individual words only",
        "answer_b": "By breaking words into subword units and assigning numerical IDs",
        "answer_c": "By storing text directly without transformation",
        "answer_d": "By removing all punctuation automatically",
        "correct_answer": "B"
    },
    {
        "question": "Which role in a conversation sets the overall behavior for a model?",
        "answer_a": "user",
        "answer_b": "system",
        "answer_c": "assistant",
        "answer_d": "developer",
        "correct_answer": "B"
    },
    {
        "question": "Which statement is TRUE about tool usage in chat templates?",
        "answer_a": "Tools cannot be used within the conversation context.",
        "answer_b": "Tools are used only for logging messages.",
        "answer_c": "Tools allow the assistant to offload tasks like web search or calculations.",
        "answer_d": "Tools are unsupported in all modern LLMs.",
        "correct_answer": "C"
    }
]