"Test yourself across the whole journey. Recalling an answer before checking is how learning turns into knowing."
Level: All levels · Time: ~15 min · Prerequisites: Modules 1–51
Learning Objectives
By the end of this module, you will be able to:
- Test your understanding across the whole path
- Spot topics worth revisiting
- Reinforce key ideas through active recall
How to Use This Quiz
Answer all 16 first, then scroll to Answers & Explanations. Each miss points to the module to revisit. The questions are deliberately spread across the whole path — foundations, LLMs, RAG, MCP, edge AI, and production — so your results form a little heat map of your strengths and gaps. Resist the urge to check answers one at a time; committing to all sixteen first gives you an honest picture and makes each recall effort do its work.
Concept: This is active recall in action — the single most effective study technique in the research. Pulling an answer out of your memory (even a wrong one) strengthens it far more than re-reading the module would. The mild discomfort of "wait, which one is it?" is the feeling of learning happening.
Try this: Decide your answer before peeking. Hesitation between two options is a signal of exactly what to review.
The Questions
1. Machine learning differs from traditional programming because it…
- A. Runs faster
- B. Learns rules from data instead of following hand-written ones
- C. Needs no data
- D. Only works in the cloud
2. "Deep" learning refers to…
- A. Deep philosophy
- B. Many hidden layers in a neural network
- C. Deep-sea data
- D. Slow training
3. What does training a neural network adjust?
- A. The screen resolution
- B. The weights
- C. The programming language
- D. The dataset size
4. A CNN is especially good at…
- A. Images
- B. Spreadsheets
- C. Audio only
- D. Nothing
5. The transformer's key idea is…
- A. Recursion
- B. Attention
- C. Compression
- D. Encryption
6. An LLM generates text by predicting…
- A. The weather
- B. The next token
- C. A random word
- D. The user's mood
7. Prompt injection is…
- A. A speed boost
- B. Hidden malicious instructions in content the model reads
- C. A prompt template
- D. A type of model
8. RAG reduces hallucinations by…
- A. Making the model bigger
- B. Retrieving real facts and answering from them
- C. Lowering temperature
- D. Removing prompts
9. An MCP server exposes tools, resources, and…
- A. Prompts
- B. Passwords
- C. Pixels
- D. Payments
10. The safest way to treat a tool's output is as…
- A. A trusted command
- B. Untrusted data, never a command
- C. A password
- D. Irrelevant
11. Edge AI runs models…
- A. Only in the cloud
- B. On local devices
- C. On paper
- D. Never
12. Quantization makes a model…
- A. Larger and slower
- B. Smaller and faster, with minor quality loss
- C. More accurate always
- D. Encrypted
13. Model "drift" means…
- A. The model moves on screen
- B. Accuracy slips as the world changes
- C. The file corrupts
- D. It speeds up
14. Batch vs. real-time prediction is chosen based on…
- A. How fresh the prediction must be
- B. The programming language
- C. The color scheme
- D. The team size
15. The biggest AI cost-saving lever is usually…
- A. A bigger model
- B. Using the smallest model that's good enough
- C. More GPUs
- D. Longer prompts
16. A great capstone project is…
- A. Fancy but unfinished
- B. Complete and end-to-end, even if simple
- C. Only a notebook
- D. Never shared
Answers & Explanations
- B — ML learns rules from data. (Module 3)
- B — Many hidden layers. (Module 4)
- B — The weights. (Module 5)
- A — Images (CNNs). (Module 6)
- B — Attention. (Module 8)
- B — The next token. (Module 12)
- B — Hidden malicious instructions in read content. (Modules 14, 28)
- B — Retrieves real facts to answer from. (Module 21)
- A — Prompts. (Module 25)
- B — Untrusted data, never a command. (Module 28)
- B — On local devices. (Module 31)
- B — Smaller/faster, minor quality loss. (Module 33)
- B — Accuracy slips as the world changes. (Module 46)
- A — How fresh the prediction must be. (Module 39)
- B — Smallest model that's good enough. (Module 49)
- B — Complete and end-to-end, even if simple. (Module 51)
How did you do?
- 13–16: Outstanding — you've internalized the whole path.
- 8–12: Strong — revisit the modules behind any misses.
- 0–7: Re-skim the group summaries (Foundations, LangChain, MCP, Edge, Production) and retry.
Whatever your score, look at which questions you missed rather than just the number. If your misses cluster in one area — say, several production questions (drift, cost, serving) — that's a clear signal to spend an evening re-reading that group, not the whole course. Scattered single misses are usually just fuzzy recall that a quick re-read fixes. And a retake a few days later, from cold memory, is worth far more than immediately re-reading the answers now: the spacing is what makes the knowledge stick for good.
Key Takeaway: These 16 questions span the entire journey — foundations, transformers and LLMs, RAG, MCP, edge AI, and production engineering. If you can answer them, you understand how the pieces of AI engineering fit together. Any misses point straight to the module worth another look.