Final Exam: Introduction to LLMs
Comprehensive assessment covering all four modules.
25 questions - Test your overall understanding of Large Language Models.
Question 1
What is the fundamental mechanism behind how LLMs generate text?
A) Looking up answers in a database
B) Predicting the next token based on patterns learned during training
C) Copying from the internet in real-time
D) Following pre-written scripts
Show Answer
Correct Answer: B
LLMs generate text by predicting the next token (word piece) based on patterns learned from training data. This simple mechanism underlies all their capabilities.
Question 2
Why did ChatGPT become the fastest-growing consumer application in history?
A) It was the first AI ever created
B) It combined capable AI with an accessible, conversational interface
C) It was mandatory for all internet users
D) It was completely free with no limits
Show Answer
Correct Answer: B
ChatGPT combined GPT-3.5's capabilities with RLHF training for natural conversation, making advanced AI accessible to everyone through a simple chat interface.
Question 3
What is the relationship between tokens and words?
A) 1 token = exactly 1 word
B) 1 token ≈ ¾ of a word on average
C) 1 token = 10 words
D) They are completely unrelated
Show Answer
Correct Answer: B
Tokens are word pieces—on average, one token equals about ¾ of a word. Longer or unusual words are split into multiple tokens.
Question 4
What does "hallucination" mean in the context of LLMs?
A) The AI experiencing dreams
B) Generating plausible-sounding but incorrect information
C) Displaying visual effects
D) Running slowly
Show Answer
Correct Answer: B
Hallucination is when an LLM generates information that sounds convincing but is factually incorrect or completely made up.
Question 5
What is a context window?
A) A pop-up help screen
B) The maximum text the model can process at once
C) A debugging tool
D) A window for viewing context menus
Show Answer
Correct Answer: B
The context window is the maximum amount of text (in tokens) that an LLM can "see" when generating a response—like its short-term memory limit.
Question 6
What company created Claude and focuses on AI safety?
A) OpenAI
B) Google
C) Anthropic
D) Meta
Show Answer
Correct Answer: C
Anthropic, founded by former OpenAI researchers, created Claude with a focus on AI safety and alignment.
Question 7
What makes Meta's Llama models significant?
A) They are the most expensive
B) They are open-source, allowing anyone to run them locally
C) They only work on Facebook
D) They are the oldest LLMs
Show Answer
Correct Answer: B
Llama models are open-source, meaning anyone can download the weights and run them locally, enabling privacy and customization.
Question 8
What is RLHF?
A) Really Large Human Files
B) Reinforcement Learning from Human Feedback
C) Rapid Language Human Formatting
D) Regular Learning High Frequency
Show Answer
Correct Answer: B
RLHF (Reinforcement Learning from Human Feedback) is a training technique where humans rate AI outputs, and the model learns to produce higher-rated responses.
Question 9
Why would output tokens cost more than input tokens?
A) They are higher quality
B) Generating output requires more computation
C) Marketing reasons
D) They take more storage
Show Answer
Correct Answer: B
Generating each output token requires running through the model's prediction process, which is more computationally intensive than processing input.
Question 10
What is the main advantage of running LLMs locally?
A) Always faster
B) Privacy—data never leaves your machine
C) Always cheaper
D) Better quality
Show Answer
Correct Answer: B
Local LLMs keep all data on your machine, ensuring privacy for sensitive applications like healthcare, legal, or financial use cases.
Question 11
What is "temperature" in LLM settings?
A) Server cooling level
B) A control for output randomness/creativity
C) Processing speed
D) Model size
Show Answer
Correct Answer: B
Temperature controls how random or deterministic outputs are. Low temperature = predictable; high temperature = more creative but riskier.
Question 12
What does "knowledge cutoff" mean?
A) When the model stops working
B) The date after which training data doesn't exist
C) A limit on what topics can be discussed
D) Maximum knowledge capacity
Show Answer
Correct Answer: B
Knowledge cutoff is the date beyond which the model has no training data. Events after this date are unknown to the model.
Question 13
What is multimodal AI?
A) AI with multiple personalities
B) AI that processes multiple data types (text, images, audio)
C) AI running on multiple computers
D) AI speaking multiple languages
Show Answer
Correct Answer: B
Multimodal AI can understand and generate multiple types of content—text, images, audio, and video—often in combination.
Question 14
What are AI agents?
A) Human AI researchers
B) AI systems that can take actions and complete tasks autonomously
C) AI sales representatives
D) Simple chatbots
Show Answer
Correct Answer: B
AI agents can take actions, use tools, plan multi-step tasks, and work autonomously to complete complex goals.
Question 15
What is quantization in the context of local LLMs?
A) Counting parameters
B) Reducing model precision to make it smaller and faster
C) Quality measurement
D) Splitting into pieces
Show Answer
Correct Answer: B
Quantization reduces the precision of model weights (e.g., from 16-bit to 4-bit), making models smaller and faster at some quality cost.
Question 16
What is an API in the context of LLMs?
A) A type of AI model
B) An interface for applications to communicate with LLM services
C) A programming language
D) An AI personality
Show Answer
Correct Answer: B
An API (Application Programming Interface) allows applications to send prompts to LLM services and receive generated responses programmatically.
Question 17
Why is GPT called "Generative Pre-trained Transformer"?
A) Random marketing name
B) Generative (creates), Pre-trained (learned before use), Transformer (architecture)
C) Named after inventors
D) Based on location
Show Answer
Correct Answer: B
GPT describes the model: Generative (creates new text), Pre-trained (trained on data before deployment), Transformer (the neural network architecture used).
Question 18
What is RAG (Retrieval-Augmented Generation)?
A) A random tag generator
B) Combining LLMs with knowledge base retrieval to reduce hallucination
C) A pricing model
D) A testing framework
Show Answer
Correct Answer: B
RAG retrieves relevant documents from a knowledge base and includes them in the prompt, grounding LLM responses in verified information.
Question 19
What makes Claude particularly good for document analysis?
A) Fastest processing
B) Very large context window (up to 200K tokens)
C) Cheapest price
D) Best images
Show Answer
Correct Answer: B
Claude's context window of up to 200K tokens allows it to process very long documents—even entire books—in a single prompt.
Question 20
What is "emergent abilities" in LLMs?
A) Emergency features
B) Capabilities that appear at scale without explicit programming
C) Add-on purchases
D) Startup abilities
Show Answer
Correct Answer: B
Emergent abilities are capabilities (like reasoning or translation) that appear when models reach sufficient scale, even though they weren't explicitly programmed.
Question 21
How do LLMs "remember" previous conversation?
A) Permanent database storage
B) The entire conversation is re-sent with each message
C) Magic
D) Cloud backup
Show Answer
Correct Answer: B
LLMs don't have persistent memory. The entire conversation history is included in each message, up to the context limit.
Question 22
What is the safety debate around open-source LLMs?
A) They're completely safe
B) Open access enables both beneficial use and potential misuse
C) Only closed models are dangerous
D) Safety isn't a concern
Show Answer
Correct Answer: B
The debate centers on whether open access enables innovation and transparency, or enables malicious use and removal of safety measures.
Question 23
When should you NOT rely solely on LLM output?
A) Creative writing
B) Medical, legal, or financial decisions
C) Email drafts
D) Brainstorming
Show Answer
Correct Answer: B
High-stakes decisions (medical, legal, financial) should never rely solely on LLM output due to hallucination risk and lack of accountability.
Question 24
What tool makes running local LLMs simple for beginners?
A) Microsoft Word
B) Ollama or LM Studio
C) Photoshop
D) Excel
Show Answer
Correct Answer: B
Ollama and LM Studio make downloading and running local LLMs simple without requiring technical expertise.
Question 25
What is the key insight about what LLMs actually do?
A) They understand like humans
B) They predict statistically likely text based on patterns
C) They copy from the internet
D) They think consciously
Show Answer
Correct Answer: B
LLMs predict statistically likely text based on patterns learned during training. This simple mechanism underlies all their capabilities, including their limitations.
Scoring Guide
| Score | Grade | Interpretation |
|---|---|---|
| 23-25 | A | Excellent understanding of LLMs |
| 20-22 | B | Good understanding, minor gaps |
| 17-19 | C | Adequate understanding, review recommended |
| 14-16 | D | Basic understanding, significant review needed |
| Below 14 | F | Re-study course materials |
What's Next?
Passed? Consider:
- Learning prompt engineering
- Exploring API development
- Diving deeper into machine learning
Need Review?
- Revisit module chapters
- Retake module quizzes
- Use the Glossary for term clarification
Congratulations!
You've completed Introduction to LLMs! You now have a solid conceptual understanding of:
✅ What LLMs are and how they work
✅ Major providers and their differences
✅ Practical considerations (APIs, tokens, costs)
✅ Limitations and how to work around them
✅ The broader ecosystem and future trends
Welcome to the world of AI!