Quiz: What Are LLMs?
Test your knowledge about Large Language Models and their fundamentals.
30 questions - Click on each question to reveal the correct answer.
Question 1
What does LLM stand for?
A) Large Learning Machine
B) Large Language Model
C) Linguistic Logic Module
D) Language Learning Mechanism
Show Answer
Correct Answer: B
LLM stands for Large Language Model. These are AI models trained on vast amounts of text data that can understand and generate human-like language.
Question 2
What is the fundamental task that language models perform?
A) Memorizing entire books
B) Predicting the next word/token
C) Storing facts in a database
D) Translating between languages
Show Answer
Correct Answer: B
Language models fundamentally predict what text comes next. Everything they do—from answering questions to writing code—is based on predicting the next word or token.
Question 3
What does GPT stand for?
A) General Purpose Technology
B) Generative Pre-trained Transformer
C) Giant Pattern Translator
D) Guided Processing Text
Show Answer
Correct Answer: B
GPT stands for Generative Pre-trained Transformer. Generative means it creates text, pre-trained means it learns from data before use, and Transformer is the architecture type.
Question 4
Why is "large" important in Large Language Models?
A) It sounds more impressive
B) Large models can only run on large computers
C) Scale enables emergent abilities like reasoning
D) Large means longer responses
Show Answer
Correct Answer: C
The scale of LLMs (billions of parameters and massive training data) enables emergent abilities—capabilities like reasoning and coding that weren't explicitly programmed but appear at scale.
Question 5
What are "parameters" in an LLM?
A) The questions you can ask
B) The adjustable values learned during training
C) The commands to control the model
D) The languages the model knows
Show Answer
Correct Answer: B
Parameters are the adjustable values (weights) in the model that are tuned during training. More parameters generally means more capacity to learn patterns.
Question 6
How many parameters did GPT-3 have?
A) 1.5 million
B) 1.5 billion
C) 175 billion
D) 1.75 trillion
Show Answer
Correct Answer: C
GPT-3 had 175 billion parameters, which was a massive leap from GPT-2's 1.5 billion. This scale enabled new capabilities like few-shot learning.
Question 7
What made ChatGPT different from GPT-3?
A) It was much larger
B) It was trained to have conversations using RLHF
C) It could generate images
D) It was open source
Show Answer
Correct Answer: B
ChatGPT used Reinforcement Learning from Human Feedback (RLHF) to fine-tune the model for natural conversations, making it much more user-friendly than the base GPT-3.
Question 8
How quickly did ChatGPT reach 100 million users?
A) 2 weeks
B) 2 months
C) 6 months
D) 1 year
Show Answer
Correct Answer: B
ChatGPT reached 100 million users in just 2 months, making it the fastest-growing consumer application in history.
Question 9
Which company created GPT and ChatGPT?
A) Google
B) Microsoft
C) OpenAI
D) Meta
Show Answer
Correct Answer: C
OpenAI created GPT and ChatGPT. Microsoft is a major investor but didn't create the models.
Question 10
Which company developed Claude?
A) OpenAI
B) Google
C) Anthropic
D) Microsoft
Show Answer
Correct Answer: C
Anthropic, founded by former OpenAI researchers, developed Claude with a focus on AI safety and alignment.
Question 11
What is Claude known for compared to other LLMs?
A) Being the fastest
B) Having the longest context window
C) Being the cheapest
D) Only working in English
Show Answer
Correct Answer: B
Claude is known for its very long context window (up to 200K tokens), allowing it to process entire books or large codebases in a single prompt.
Question 12
What does "emergent abilities" mean for LLMs?
A) Abilities that need to be purchased separately
B) Capabilities that appear at scale without explicit training
C) Emergency backup features
D) Abilities that only work when emerging from sleep mode
Show Answer
Correct Answer: B
Emergent abilities are capabilities that weren't explicitly programmed but "emerge" when models reach sufficient scale—like reasoning, coding, and translation.
Question 13
What is Google's main LLM called?
A) Bard
B) Gemini
C) Lambda
D) Pixel AI
Show Answer
Correct Answer: B
Google's main LLM is Gemini. Bard was the earlier name for the consumer product, which was rebranded to Gemini.
Question 14
Which company released Llama as open source?
A) OpenAI
B) Google
C) Meta
D) Microsoft
Show Answer
Correct Answer: C
Meta released Llama as open source (with some conditions), allowing researchers and companies to use and modify the models freely.
Question 15
What is one major advantage of open-source LLMs like Llama?
A) They're always more powerful
B) You can run them locally for privacy
C) They never make mistakes
D) They don't require any computing power
Show Answer
Correct Answer: B
Open-source LLMs can run locally on your own hardware, meaning your data never leaves your machine—important for privacy-sensitive applications.
Question 16
What year was the Transformer architecture introduced?
A) 2015
B) 2017
C) 2019
D) 2022
Show Answer
Correct Answer: B
The Transformer architecture was introduced in 2017 in the paper "Attention Is All You Need." It became the foundation for all modern LLMs.
Question 17
Why did OpenAI initially hesitate to release GPT-2?
A) It was too slow
B) Concerns about misuse for fake news
C) It didn't work well
D) Legal issues
Show Answer
Correct Answer: B
OpenAI was concerned that GPT-2's ability to generate convincing fake text could be misused for disinformation and fake news.
Question 18
What is "few-shot learning" in the context of GPT-3?
A) Learning from a few mistakes
B) Performing tasks with just a few examples in the prompt
C) Training with limited data
D) Taking few shots to generate an answer
Show Answer
Correct Answer: B
Few-shot learning means the model can understand a task from just a few examples provided in the prompt, without any additional training.
Question 19
What new capability did GPT-4 add compared to GPT-3?
A) Voice generation
B) Vision (understanding images)
C) Physical robotics
D) Smell detection
Show Answer
Correct Answer: B
GPT-4 added multimodal capabilities, specifically the ability to understand and reason about images alongside text.
Question 20
What is Microsoft's AI coding assistant called?
A) Microsoft Coder
B) Azure AI
C) GitHub Copilot
D) Visual Studio Bot
Show Answer
Correct Answer: C
GitHub Copilot is Microsoft's AI coding assistant, powered by OpenAI models, integrated into code editors.
Question 21
How do LLMs "know" facts like "The capital of France is Paris"?
A) They have a database of facts
B) They've learned patterns associating those words
C) They access the internet in real-time
D) Programmers manually input facts
Show Answer
Correct Answer: B
LLMs don't store facts in a database. They've learned patterns from training data that strongly associate "capital of France" with "Paris."
Question 22
What is a "context window" in LLMs?
A) The graphical user interface
B) The maximum amount of text the model can process at once
C) A window that shows the model's thinking
D) The time limit for responses
Show Answer
Correct Answer: B
The context window is the maximum amount of text (tokens) the model can "see" and consider when generating a response.
Question 23
What company is Claude's creator, Anthropic, known for focusing on?
A) Speed and efficiency
B) AI safety and alignment
C) Gaming applications
D) Social media integration
Show Answer
Correct Answer: B
Anthropic was founded with a focus on AI safety and alignment—ensuring AI systems behave according to human values.
Question 24
What is the relationship between Microsoft and OpenAI?
A) They are competitors
B) Microsoft is a major investor in OpenAI
C) OpenAI owns Microsoft
D) No relationship
Show Answer
Correct Answer: B
Microsoft is a major investor in OpenAI and has exclusive cloud partnership rights, integrating OpenAI's technology into Microsoft products.
Question 25
Which LLM would be best for analyzing a 500-page legal document?
A) GPT-3.5 (short context)
B) Claude (very long context)
C) Any model—they're all the same
D) Llama (smallest size)
Show Answer
Correct Answer: B
Claude's very long context window (up to 200K tokens) makes it ideal for analyzing very long documents that wouldn't fit in other models' context windows.
Question 26
What is "RLHF" that was used to train ChatGPT?
A) Really Large Human Files
B) Reinforcement Learning from Human Feedback
C) Rapid Language Human Formatting
D) Real-time Learning for Higher Functions
Show Answer
Correct Answer: B
RLHF (Reinforcement Learning from Human Feedback) is a training technique where humans rate model outputs, and the model learns to produce higher-rated responses.
Question 27
What does "multimodal" mean for an LLM like GPT-4?
A) It can run on multiple computers
B) It can process different types of data (text, images, etc.)
C) It has multiple personalities
D) It speaks multiple languages
Show Answer
Correct Answer: B
Multimodal means the model can understand and work with multiple types of data—like text, images, and potentially audio and video.
Question 28
Which statement about LLM training data is TRUE?
A) LLMs are trained only on Wikipedia
B) Training data includes web pages, books, code, and more
C) Training data is secret and never discussed
D) LLMs don't need training data
Show Answer
Correct Answer: B
Modern LLMs are trained on diverse datasets including web pages, books, scientific papers, code repositories, and more.
Question 29
What is one advantage of closed-source LLMs (like GPT-4) over open-source?
A) They're always free
B) More polished user experience and regular updates
C) You can modify the model code
D) Better privacy
Show Answer
Correct Answer: B
Closed-source LLMs typically offer more polished user experiences, regular updates, and customer support—though you give up control and customization.
Question 30
Based on what you've learned, why can LLMs sometimes give incorrect information?
A) They're programmed to lie occasionally
B) They predict likely text based on patterns, not verified facts
C) The internet has no correct information
D) They only work offline
Show Answer
Correct Answer: B
LLMs predict what text is likely based on patterns they've learned. If a pattern exists but is incorrect, or if the model generates plausible-sounding but wrong information, errors occur.
Results
| Score | Level |
|---|---|
| 27-30 | Excellent - Strong understanding of LLM fundamentals |
| 21-26 | Good - Some review may be helpful |
| 15-20 | Average - Review the chapters again |
| Below 15 | Needs comprehensive review |