Quiz: Module 4 - Major LLM Families

30 questions covering GPT, Claude, and open-source models.

---

Question 1

Which company developed the GPT series?

1. OpenAI
2. Anthropic
3. Google
4. Meta

Question 2

What was GPT-3's key innovation?

1. Multimodal capabilities
2. In-context learning at scale
3. Constitutional AI
4. Open source release

Question 3

What technique made ChatGPT more helpful than GPT-3?

1. More parameters
2. Longer context
3. RLHF (Reinforcement Learning from Human Feedback)
4. Better tokenizer

Question 4

Who founded Anthropic?

1. Mark Zuckerberg
2. Former OpenAI researchers
3. Google engineers
4. Meta AI team

Question 5

What is Constitutional AI?

1. A government regulation
2. Training AI with a set of principles instead of case-by-case labeling
3. A model architecture
4. A tokenization method

Question 6

What is Claude 3's context window size?

1. 4K tokens
2. 32K tokens
3. 100K tokens
4. 200K tokens

Question 7

Which company released LLaMA?

1. OpenAI
2. Google
3. Meta
4. Anthropic

Question 8

What is Mixtral?

1. A closed-source model
2. Mistral's Mixture of Experts model
3. A training framework
4. A tokenizer

Question 9

What license does LLaMA 2 use?

1. Research only
2. Commercial use allowed
3. GPL
4. Proprietary

Question 10

What's the main advantage of open-weight models?

1. Always better performance
2. Full control, local deployment, custom fine-tuning
3. Easier to use
4. No training needed

Question 11

Approximately how many parameters does GPT-4 have?

1. 175 billion
2. 500 billion
3. ~1.8 trillion (estimated)
4. 10 trillion

Question 12

What capability did GPT-4 add that GPT-3 lacked?

1. Text generation
2. Image understanding (multimodal)
3. Translation
4. Summarization

Question 13

Which Claude model is fastest and most cost-effective?

1. Claude 3 Haiku
2. Claude 3 Sonnet
3. Claude 3 Opus
4. Claude 2

Question 14

What is Sliding Window Attention used by Mistral?

1. Attention only over a local window of tokens for efficiency
2. Full attention over all tokens
3. Attention that moves across the sequence
4. Cross-attention between models

Question 15

What year was GPT-1 released?

1. 2016
2. 2017
3. 2018
4. 2019

Question 16

What is a key focus of Anthropic's research?

1. Maximum model size
2. AI safety and alignment
3. Speed optimization
4. Hardware efficiency

Question 17

What does "open weights" mean?

1. Model weights are publicly available
2. Training code is open
3. Training data is public
4. API is free

Question 18

Which platform hosts most open-source LLMs?

1. GitHub
2. Hugging Face
3. Kaggle
4. TensorFlow Hub

Question 19

What is GPT's architectural type?

1. Encoder-only
2. Encoder-decoder
3. Decoder-only
4. Hybrid

Question 20

What makes Claude "honest" according to Anthropic?

1. It never makes mistakes
2. It acknowledges uncertainty and limitations
3. It only uses verified facts
4. It has access to the internet

Question 21

What is the typical size of Mistral 7B?

1. 7 billion parameters
2. 7 million parameters
3. 70 billion parameters
4. 700 million parameters

Question 22

Why was GPT-2's release initially limited?

1. Technical issues
2. Concerns about misuse potential
3. Commercial reasons
4. Licensing disputes

Question 23

What is Grouped Query Attention?

1. Sharing key-value heads across multiple query heads for efficiency
2. Grouping queries by topic
3. Sequential query processing
4. Query caching

Question 24

Which Claude 3 model is best for complex tasks?

1. Claude 3 Haiku
2. Claude 3 Sonnet
3. Claude 3 Opus
4. All are equal

Question 25

What is LLaMA 3's improvement over LLaMA 2?

1. Smaller size
2. Better performance on benchmarks
3. Different architecture
4. New license

Question 26

What does MoE stand for in LLM context?

1. Model of Everything
2. Mixture of Experts
3. Method of Encoding
4. Module of Execution

Question 27

How do closed models like GPT-4 typically charge?

1. Monthly subscription
2. Per token (input and output)
3. Per query
4. Free

Question 28

What is Code Llama?

1. LLaMA fine-tuned specifically for code
2. A coding IDE
3. A code formatter
4. A debugging tool

Question 29

Which company is behind Mistral?

1. OpenAI
2. Meta
3. Mistral AI (French startup)
4. Google

Question 30

What is the benefit of Mixtral's sparse MoE?

1. Large capacity with lower inference cost (only subset of experts active)
2. Simpler architecture
3. Better training stability
4. Smaller model size

---

Quiz Complete!

Passing Score: 24/30 (80%)

← Chapter 3Module 5: Tokenization →