GenAI & LLM Engineering Objective Questions and Answers - Set 2

This GenAI & LLM engineering quiz contains carefully curated objective questions with correct answers and clear explanations. It is designed for developers and ML engineers to test your skills in building, optimizing, evaluating, and deploying LLM-powered applications. This is part 2 of the series.

Practice GenAI & LLM Engineering MCQs with Detailed Explanations

Answer at least 12 questions to submit and verify answers.

16 What is the primary benefit of using caching in LLM APIs? Medium

Improves training accuracy

Reduces repeated inference cost and latency

Improves embedding quality

Increases randomness

17 Which approach is best for grounding LLM responses in real-time data? Medium

Offline fine-tuning only

RAG with live data sources

Static prompts

Higher temperature sampling

18 What is a major challenge when deploying LLM agents in production? Hard

Tokenization

Tool misuse and uncontrolled actions

Embedding dimension limits

Vector compression

19 Which evaluation method best measures factual consistency of LLM outputs? Hard

Human evaluation

Perplexity only

Token count

Latency measurement

20 What does top-k sampling control? Medium

Number of tokens generated

Vocabulary size considered at each step

Embedding dimension

Batch size

21 Which technique helps align LLM outputs with human preferences? Medium

RLHF

Unsupervised pretraining only

Token pruning

Quantization

22 What is the main benefit of using streaming responses in chat applications? Medium

Lower model size

Improved perceived latency

Better embeddings

Reduced hallucinations

23 Which risk increases when using very large overlapping chunks in RAG? Hard

Loss of context

Higher retrieval noise and redundancy

Lower recall

Embedding collapse

24 What is the main purpose of prompt templates? Medium

Standardize instructions and structure inputs

Train the LLM

Reduce vector dimensions

Compress responses

25 Which failure is most likely if the retriever returns irrelevant documents? Hard

Overfitting

Grounded hallucination

Prompt injection

Mode collapse

26 What is the role of temperature = 0 in decoding? Easy

Random sampling

Deterministic output

Maximum diversity

Unbounded randomness

27 Which component is critical for semantic search at scale? Medium

Vector database

Beam search

Tokenizer

Dropout

28 What is the key advantage of hybrid search (keyword + vector)? Hard

Lower memory usage

Better recall and precision across diverse queries

Faster tokenization

Lower training cost

29 Which is a common cause of prompt overfitting in production? Hard

Highly rigid templates that fail on edge cases

Large context windows

Quantized models

Low batch size

30 What is the main goal of model distillation for LLMs? Medium

Increase model size

Transfer knowledge to a smaller model

Increase context length

Improve training data quality