Quiz Time: Building and Using

Now six questions about using, and building with A I. As before, I will read the question, and the four answers. Take a few seconds to think, and then I will reveal the answer.

• Hallucinations, RAG, chunks
• Agents, fine-tuning, security
• Pause, think, then check

Hallucination

Question 1. What is a hallucination? A, A fast answer. B, A confident answer that is false. C, A very short answer. D, A translated answer. Pause the video, and think. The answer is B. Models predict plausible text, so they can sound sure while being wrong.

RAG

Question 2. What does R A G add to a model? A, A bigger screen. B, Relevant documents found at question time. C, More weights. D, A new voice. Pause the video, and think. The answer is B. It retrieves the right chunks, and adds them to the prompt, so answers are grounded.

Chunking

Question 3. Why do we split documents into chunks? A, So each piece has a clear meaning for search. B, To make files bigger. C, To hide the text. D, To save electricity. Pause the video, and think. The answer is A. Smaller pieces mean a question can find the exact paragraph that answers it.

Agents

Question 4. What is the agent loop? A, Think, act, observe, repeat. B, Read, delete, save, quit. C, Train, sleep, wake, train. D, Draw, paint, print, stop. Pause the video, and think. The answer is A. The agent plans, calls a tool, reads the result, and decides whether to continue.

Fine-tuning

Question 5. When is fine tuning the best choice? A, To add changing facts. B, To teach a stable style or skill. C, To make prompts shorter. D, To avoid testing. Pause the video, and think. The answer is B. Facts change and are better retrieved. Behaviour is stable and is better learned.

Security

Question 6. What is prompt injection? A, A medical shot. B, Hidden instructions placed inside content the model reads. C, A faster API. D, A kind of chart. Pause the video, and think. The answer is B. Treat every document and web page as untrusted, and limit what tools can do.

Summary