Free practice exam
Large Language Model Engineering (LLM) practice questions
Practise large language model fundamentals: tokens and embeddings, transformers, training and fine-tuning, evaluation, retrieval, safety and cost.
These 15 questions are separate from the real LLM capstone, so you can see every answer here. Pick an option to check yourself, and use the chapter link to revise.
Score: 0 / 0 answered · 15 questions
1. What is a token in a language model?
Answer: B. A unit of text, such as a word piece, that the model processes
Text is split into tokens; limits and costs are counted in tokens.
Study this in chapter 22. What does an embedding represent?
Answer: B. A vector of numbers capturing the meaning of text so similar items sit close together
Embeddings power semantic search and retrieval.
Study this in chapter 23. What is the key mechanism introduced by the transformer architecture?
Answer: B. Self-attention, letting each token weigh other tokens in the sequence
'Attention Is All You Need' (Vaswani et al., 2017) introduced the transformer.
Study this in chapter 34. What is a typical pre-training objective for decoder-only LLMs?
Answer: B. Predicting the next token
Next-token prediction over large text corpora is the standard objective.
Study this in chapter 45. What is LoRA?
Answer: B. A parameter-efficient fine-tuning method that trains small low-rank adapter matrices
LoRA cuts memory and cost by training far fewer parameters.
Study this in chapter 56. Why can benchmark scores be misleading?
Answer: B. Test data may have leaked into training, and benchmarks may not match your real task
Contamination and mismatch mean you should also evaluate on your own data.
Study this in chapter 67. What does quantisation do?
Answer: B. Stores model weights at lower precision to reduce memory and speed up inference
Lower precision trades a little accuracy for big efficiency gains.
Study this in chapter 78. What problem is retrieval-augmented generation (RAG) mainly designed to reduce?
Answer: B. Answers without grounding, by supplying relevant source documents at query time
RAG grounds answers in retrieved text and allows citations; it reduces but does not eliminate hallucination.
Study this in chapter 109. In RAG, what is 'chunking'?
Answer: B. Splitting documents into smaller passages for embedding and retrieval
Chunk size affects retrieval quality.
Study this in chapter 1010. What is a prompt-injection attack?
Answer: B. Malicious text that tries to override the model's instructions, often hidden in content it reads
Untrusted content must be treated as data, not instructions.
Study this in chapter 1111. What is the purpose of a system prompt?
Answer: B. Setting the model's role, rules and constraints for a conversation
System instructions frame behaviour; they are not a security boundary on their own.
Study this in chapter 1712. In agent systems, what is 'tool calling'?
Answer: B. The model requesting a structured call to an external function or API and using the result
Tools let models fetch data or take actions; permissions should be limited.
Study this in chapter 1813. What does RLHF stand for?
Answer: B. Reinforcement learning from human feedback
Human preference data is used to align model outputs.
Study this in chapter 1914. How does DPO differ from classic RLHF?
Answer: B. It optimises directly on preference pairs without training a separate reward model
Direct Preference Optimization (Rafailov et al., 2023) simplifies alignment training.
Study this in chapter 1915. Which change usually lowers LLM serving cost most directly?
Answer: B. Shorter prompts, caching, and routing simple tasks to smaller models
Cost scales with tokens and model size.
Study this in chapter 22
Ready for the real LLM capstone?
The final uses different questions drawn from the same chapters. Track your attempts and study material in your course portal.
Authoritative sources
Independent primary sources used to check and corroborate the guidance on this page.
NIST National Initiative for Cybersecurity Education (NICE)US national framework for cybersecurity work roles and competencies.
Financial Action Task Force (FATF)Global AML/CFT standards referenced by financial-crime curricula.
FBI Internet Crime Complaint Center — Annual ReportsPrimary loss and typology data used in fraud-investigation training.
US Federal Trade Commission — Consumer Sentinel Data BookOfficial complaint statistics for teaching fraud trend analysis.