LLM & Prompting Quiz — Quiz

Question 1 of 10

LLMs generate text by predicting…

aThe next tokenbThe next sentencecThe whole answer at oncedKeywords

Question 2 of 10

RAG stands for…

aRandom Answer GenerationbRecursive Attention GraphcRapid AI GatewaydRetrieval-Augmented Generation

Question 3 of 10

Lower temperature makes output…

aMore randombLongercMore deterministicdShorter

Question 4 of 10

Embeddings are…

aCompressed imagesbNumeric vectors representing meaningcEncrypted textdModel weights

Question 5 of 10

An assistant confidently states a refund policy that does not exist. This is called…

aA hallucinationbA timeoutcOverfittingdA rate limit

Question 6 of 10

What is the system message for?

messages = [
    {'role': 'system', 'content': 'You are a concise support agent for a bakery.'},
    {'role': 'user', 'content': 'Do you deliver on Sundays?'},
]

aIt is displayed to the end userbIt stores the model's answercIt is billed as free tokensdIt sets the role and rules the model should follow for the conversation

Question 7 of 10

In a RAG pipeline, why is a long document split into chunks before embedding?

aChunks make the embedding files smallerbSo retrieval returns small, focused passages that fit in the promptcBecause models cannot embed PDFsdTo randomise the order of the text

Question 8 of 10

Which prompting technique does this use?

Classify the sentiment.
Review: 'Fast delivery, great packaging' -> positive
Review: 'Item arrived broken' -> negative
Review: 'Exactly as described' ->

aZero-shot promptingbFine-tuningcFew-shot promptingdChain-of-thought prompting

Question 9 of 10

This sometimes raises a JSONDecodeError because the model adds a sentence before the JSON. Best fix?

data = json.loads(response.text)

aAsk for a defined JSON schema using the provider's structured-output mode, then validate and retry on failurebRaise the temperaturecStrip everything before the first '{' and hope for the bestdSwitch to a larger context window

Question 10 of 10

Even with temperature set to 0, the same prompt can return slightly different answers. Why?

aTemperature 0 is ignored by every providerbServing-side factors such as batching, hardware floating-point differences and silent model updates make output non-deterministiccThe prompt is re-tokenised at randomdBecause embeddings expire