Quiz: LLMs in Practice
Test your knowledge about using LLMs in real-world applications.
30 questions - Click on each question to reveal the correct answer.
Question 1
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) A security feature
Show Answer
Correct Answer: B
An API (Application Programming Interface) allows applications to communicate with LLM services, sending requests and receiving responses programmatically.
Question 2
How is a token related to words?
A) 1 token = exactly 1 word
B) 1 token ≈ ¾ of a word on average
C) 1 token = 10 words
D) Tokens and words are unrelated
Show Answer
Correct Answer: B
One token is approximately ¾ of a word on average. Words are often split into multiple tokens, especially longer or less common words.
Question 3
Why do output tokens cost more than input tokens?
A) They're higher quality
B) Generation requires more computation than reading
C) Marketing decision
D) They take up more storage
Show Answer
Correct Answer: B
Generating output requires more computational resources than processing input, as the model must run through its prediction process for each token.
Question 4
What is "hallucination" in LLMs?
A) A visual bug in the interface
B) Generating plausible but false information
C) A type of dream state
D) A feature for creative writing
Show Answer
Correct Answer: B
Hallucination is when an LLM generates information that sounds confident and plausible but is factually incorrect or made up.
Question 5
What is a "knowledge cutoff"?
A) A security feature that limits knowledge
B) The date after which the model has no training data
C) When the model stops responding
D) A limit on token usage
Show Answer
Correct Answer: B
The knowledge cutoff is the date beyond which the model has no information, as its training data only extends to that point.
Question 6
Which model would be most cost-effective for simple customer service questions?
A) GPT-4 Opus
B) Claude 3 Opus
C) GPT-3.5-turbo or Claude Haiku
D) The most expensive available
Show Answer
Correct Answer: C
For simple tasks, cheaper models like GPT-3.5-turbo or Claude Haiku are usually sufficient and much more cost-effective than premium models.
Question 7
What is a rate limit?
A) How fast the model thinks
B) Limits on requests or tokens per time period
C) The model's opinion score
D) A quality rating
Show Answer
Correct Answer: B
Rate limits restrict how many API requests or tokens you can use within a time period (per minute, hour, or day).
Question 8
What is RAG (Retrieval-Augmented Generation)?
A) A type of LLM model
B) Connecting LLMs to knowledge bases 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 the LLM's response in verified information.
Question 9
Why might an LLM give incorrect information about a 2024 event?
A) It's programmed to lie
B) The event may be after its knowledge cutoff
C) It doesn't process current events
D) 2024 is too far in the future
Show Answer
Correct Answer: B
If the event occurred after the model's training data cutoff, the model has no information about it and may hallucinate incorrect details.
Question 10
What distinguishes ChatGPT from the OpenAI API?
A) ChatGPT is for consumers; API is for developers
B) They use different AI models
C) ChatGPT is free; API is always paid
D) No difference
Show Answer
Correct Answer: A
ChatGPT is a consumer product with a chat interface, while the API is designed for developers to integrate LLM capabilities into applications.
Question 11
How can you reduce API costs?
A) Use the most expensive model
B) Write longer prompts
C) Optimize prompts, choose appropriate models, use caching
D) Make more requests
Show Answer
Correct Answer: C
Cost optimization includes using concise prompts, choosing the right model for the task, caching frequent responses, and limiting output length.
Question 12
What type of task should you NOT rely solely on LLMs for?
A) Creative writing brainstorming
B) Medical diagnosis
C) Email drafts
D) Code explanation
Show Answer
Correct Answer: B
Medical diagnosis has high stakes if wrong—LLM outputs should never be the sole basis for medical decisions without professional verification.
Question 13
What is "temperature" in API calls?
A) The heat of the server
B) A setting controlling output randomness
C) The speed of generation
D) A pricing tier
Show Answer
Correct Answer: B
Temperature controls how random or deterministic the output is. Low temperature = predictable; high temperature = more varied and creative.
Question 14
Approximately how many tokens is one page of text?
A) 50-100
B) 500-750
C) 5,000-7,500
D) 50,000
Show Answer
Correct Answer: B
A typical page of text is approximately 500-750 tokens, depending on the content.
Question 15
What is an API key?
A) A physical keyboard shortcut
B) A credential for authenticating API requests
C) A model name
D) A pricing plan
Show Answer
Correct Answer: B
An API key is a secret credential that identifies and authenticates your application when making API requests.
Question 16
Why do LLMs sound confident even when wrong?
A) They're designed to lie
B) They predict probable-sounding text regardless of accuracy
C) They're always confident
D) They know they're right
Show Answer
Correct Answer: B
LLMs generate probable-sounding text based on patterns, not truth. Incorrect answers often sound just as confident as correct ones.
Question 17
What is streaming in API responses?
A) Video generation
B) Receiving response tokens as they're generated
C) Continuous billing
D) Live updates to the model
Show Answer
Correct Answer: B
Streaming allows you to receive and display tokens as they're generated, rather than waiting for the complete response.
Question 18
What should you do before trusting an LLM-generated citation?
A) Assume it's correct
B) Verify it exists and is accurate
C) Only use it if it sounds professional
D) Check if the model is expensive
Show Answer
Correct Answer: B
LLMs often hallucinate citations that don't exist. Always verify that cited sources are real and accurately quoted.
Question 19
What advantage does Claude have for document analysis?
A) Cheapest price
B) Very large context window (200K tokens)
C) Fastest speed
D) Best at math
Show Answer
Correct Answer: B
Claude's context window can handle up to 200K tokens, allowing it to process very long documents in a single prompt.
Question 20
What is a "no-code" way to use LLM APIs?
A) Using command line
B) Tools like Zapier that don't require programming
C) There is no way without code
D) Copying and pasting code
Show Answer
Correct Answer: B
Tools like Zapier, Make, and ChatGPT's GPT Builder allow integration with LLMs without writing code.
Question 21
What might cause a "429" error from an API?
A) Server crash
B) Hitting rate limits
C) Wrong password
D) Model not found
Show Answer
Correct Answer: B
A 429 error typically means "Too Many Requests"—you've hit the rate limit and need to wait before making more requests.
Question 22
Why might open-source models be preferred for some applications?
A) They're always better
B) Privacy (data stays local) and no API costs
C) They're newer
D) They're easier to use
Show Answer
Correct Answer: B
Open-source models can run locally, keeping data private and eliminating ongoing API costs (though hosting costs apply).
Question 23
What is the "stochastic parrot" critique of LLMs?
A) They make random noises
B) They repeat patterns without true understanding
C) They're too colorful
D) They can only work in pairs
Show Answer
Correct Answer: B
The "stochastic parrot" critique suggests LLMs merely repeat patterns from training data with statistical variation, without genuine understanding.
Question 24
What is max_tokens in an API call?
A) Maximum model size
B) Maximum length of the response
C) Maximum billing amount
D) Maximum conversation length
Show Answer
Correct Answer: B
max_tokens limits how long the model's response can be, helping control costs and response length.
Question 25
How does caching help reduce API costs?
A) It makes the model smarter
B) It stores and reuses responses to repeated queries
C) It compresses tokens
D) It reduces model size
Show Answer
Correct Answer: B
Caching stores responses to common queries, so they can be reused without making new (paid) API calls.
Question 26
What types of reasoning do LLMs struggle with?
A) Simple questions
B) Complex math and multi-step logic
C) Translation
D) Summarization
Show Answer
Correct Answer: B
LLMs often struggle with complex mathematical calculations and multi-step logical reasoning tasks.
Question 27
What is the difference between input and output token pricing?
A) No difference
B) Output usually costs more
C) Input always costs more
D) Only output is charged
Show Answer
Correct Answer: B
Output tokens typically cost 2-5x more than input tokens because generation requires more computation.
Question 28
What should developers add to applications using LLMs for important tasks?
A) More AI models
B) Human review and verification
C) Faster processing
D) More tokens
Show Answer
Correct Answer: B
For important or high-stakes tasks, human review of LLM outputs helps catch errors and hallucinations.
Question 29
What is a system prompt?
A) The first user message
B) Hidden instructions that guide AI behavior
C) An error message
D) The model name
Show Answer
Correct Answer: B
A system prompt contains instructions (often hidden from users) that shape how the AI should behave throughout the conversation.
Question 30
Why is it dangerous that LLMs can generate fake citations?
A) It's not dangerous
B) Users may trust and spread false academic claims
C) It uses too many tokens
D) It slows down the model
Show Answer
Correct Answer: B
Fake citations can lead users to believe false information is academically verified, potentially spreading misinformation.
Results
| Score | Level |
|---|---|
| 27-30 | Excellent - Strong understanding of practical LLM use |
| 21-26 | Good - Some review may be helpful |
| 15-20 | Average - Review the chapters again |
| Below 15 | Needs comprehensive review |