Final exam: AI foundations: what it is, how it works, where it fails

Read in

1. A friend says, “The AI wrote it so smoothly, it must be true.” What is the best reply?
Agreed, a smooth style shows the writer researched carefully.Smooth text from a modern tool is verified before it appears.Only answers that are short or vague can contain errors.Smooth style says little about accuracy, so check key facts.
2. Which description of a large language model is most accurate?
A database that stores a verified answer for every questionA program that follows fixed rules written by expertsA search engine that retrieves and ranks pages on the internetA system that predicts likely next text from learned patterns
3. A user asks for sources on a rare topic and the tool gives five article titles. What is the wisest next step?
Keep only the title that sounds the most relevant to youLook each up to confirm it exists and says what is claimedCite all five, since article titles from AI tools are usually accurateAsk the tool to confirm that all five sources are real
4. In a very long chat, the tool stops following an instruction you gave at the start. What is the likely cause and best fix?
The tool needs a faster connection, so switch networks and try againInstructions only work for the first message, so repeat them in every messageThe context window is limited, so restate the instruction or start fresh with a summaryThe tool has lost your account settings, so sign in again to reset it
5. You ask, “Describe a typical engineer,” and every example is a young man from a wealthy country. What is the best response?
Accept it, since the tool reflects the average engineerRepeat the question until it names a different groupAsk the tool to remove every person from the exampleName the context you want and ask for varied examples
6. Which task is the riskiest to hand to a chat assistant without a check?
Turning meeting notes into a list of action itemsBrainstorming ten possible names for a school fundraiserRewriting your own paragraph to sound friendlierGiving an exact legal deadline you plan to rely on
7. The same question gets slightly different answers in two chats. What does this show?
The tool is faulty and should be reported to supportOne answer is secretly correct and the other is wrongSome randomness is normal, so one answer is a sampleThe second question was phrased more politely than the first
8. Which prompt is most likely to surface weak spots in an answer?
“Tell me which parts you are least sure about and what to verify.”“Make it sound more professional and more polished.”“Rewrite the answer in a more confident and direct tone.”“Give me the same answer again, laid out as a table.”
9. Why is an AI search tool with sources usually better for current-rule questions than a plain chat reply?
It stores the latest rules permanently so they never changeIt can point to pages you can open and check, though it may misread themIt checks every claim against official records before it shows you a replyIt gives the same wording as the original page, never a summary
10. A volunteer wants to paste a private list of names and addresses into a consumer chat tool to format it. What is the main concern?
Personal data could leak or breach policy, so use invented data or an approved toolChat tools sort names into the wrong order and mix up the addressesFormatting a long list takes the tool too long and may time outThe tool will quietly remove the addresses before it replies to you

Name to print on your certificate

We don't check identity, and we don't store your name: it lives only inside your certificate link.

Answer all 10 questions and enter a name to submit.