MySmartCoPilot
Ask Passes Your pass 7 days left 1 day left Last day

Your country

Tools that support it use your country for local currency, number formats, units and paper size. Your choice is saved only in this browser.

AI Literacy & Prompting Module 2

What AI does well and where it fails interview questions

Interview questions on what AI does well and where it fails: warm-ups, then what each harder question tests and how a good answer starts.

12 questions: 4 warm-ups and 8 with model answers · Levels: Fresher to senior

Warm-up questions

Short questions with short answers, each answered in full.

  1. Fresher to mid

    Which tasks can you usually hand to an AI tool, and which need close checking?

    Answered in the lesson What AI does well and where it slips

  2. Fresher to mid

    What is an AI hallucination, and why does NIST call it confabulation?

    Answered in the lesson Hallucinations: why AI makes things up

  3. Fresher to mid

    What is sycophancy in an AI assistant, and how can you tell it is happening?

    Answered in the lesson Sycophancy: when AI agrees too easily

  4. Fresher to mid

    What does a reasoning mode do differently from a standard chat mode, and when is it worth using?

    Answered in the lesson Reasoning models: what 'thinking' changes

Questions with model answers

What each question tests, the first step of a good answer and the follow-ups to expect.

  1. Fresher to mid

    A colleague used a chat app to total 40 invoice amounts for a vendor payment, and the total was wrong. Explain why such slips happen, and set a rule for when your team may use AI with numbers.

    What it tests
    Why exact arithmetic slips in a model that predicts tokens, how a code tool differs from the model, and checks that fit numbers.
    First step
    Say what the model did: unless the app ran code, it wrote a likely-looking total token by token without adding anything. Recompute the total in a spreadsheet and compare it with the payment before anything is sent, then look at how the slip got past the team.
    Follow-ups

    From the lesson What AI does well and where it slips

    Opening the model answers with your pass…

    Model answers open with any MySmartCoPilot pass.

    • Pro pass · 1 week₹79$2.99
    • 30 days₹179$5.99
    • 1 year₹1,699$45

    One payment; a pass never renews. The price shown is the total you pay: no fees or charges are added at checkout.

    Choose a passI have a licence key

    It also opens every other in-depth part in Learn and every Pro tool. Sales are final, except where the law gives you a right to withdraw, for example in the EU and the UK (refund policy).

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  2. Mid to senior

    Your college wants a chatbot to answer students' questions about fees, deadlines and bus timings. Which questions would you let it answer, and how would you keep its answers correct?

    What it tests
    Recognising rare, recent and local facts as weak spots, answering from the college's own current documents, and keeping those answers up to date.
    First step
    Sort the questions by how they fail. Fees, deadlines and timings are local, change every term and appear in few documents, so the bot should answer them only from the college's own current notices, quote and link the notice, and say so when it has none.
    Follow-ups

    From the lesson What AI does well and where it slips

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  3. Fresher to senior

    A junior colleague's draft legal note cites three judgments found with a chat app, and one of them cannot be found anywhere. What do you do with the note, and what checking rule do you set for citations from now on?

    What it tests
    The three forms of fake sources, why existence alone is not enough, and a citation check that a team will actually follow.
    First step
    Stop the note and treat all three citations as unverified, not just the missing one. Find each judgment in an official or trusted law report, read the paragraph cited, and check every quotation word for word; the two that exist can still be misrepresented or misquoted.
    Follow-ups

    From the lesson Hallucinations: why AI makes things up

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  4. Mid to senior

    Your support assistant answers customers from your product manuals, yet it sometimes describes features the product does not have. How do you make such answers rarer, and how do you measure whether your changes worked?

    What it tests
    Grounding answers in documents, allowing "I don't know", quotes that can be checked, and measuring hallucination on a fixed test set rather than by impression.
    First step
    Collect the wrong answers first and sort them: features from no manual at all, features of another product or version, and quotes the manuals do not contain. Each kind points to its own fix, and the collection becomes the start of a test set.
    Follow-ups

    From the lesson Hallucinations: why AI makes things up

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  5. Fresher to mid

    A manager's AI assistant called her project proposal "excellent", but the review panel rejected it for missing costs and risks. Why might the assistant have praised it, and how would you get honest feedback on the next draft?

    What it tests
    Feedback sycophancy and the cues that cause it, and prompts that ask for criticism against criteria rather than for praise.
    First step
    Look at how the feedback was asked for. If the request said she had worked hard on it or was pleased with it, the assistant had a cue to praise. Then rewrite the request around the panel's own criteria, costs and risks among them, and ask for the weakest parts first.
    Follow-ups

    From the lesson Sycophancy: when AI agrees too easily

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  6. Mid to senior

    In a planning meeting, a team lead pushed back on an AI tool's demand forecast, and the tool changed its number at once. How should the team treat the two numbers, and what process would you put in place?

    What it tests
    Recognising sycophancy under pushback, why a changed answer is not evidence, and a team process that settles disputed numbers outside the chat.
    First step
    Treat both numbers as unchecked: the second answers the lead's pushback, not new data. Ask what the forecast was built from, recompute it from that data with a method the team can inspect, and decide only on the recomputed figure.
    Follow-ups

    From the lesson Sycophancy: when AI agrees too easily

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  7. Mid to senior

    A teammate accepts an AI tool's eligibility decision on a loan application because its displayed reasoning "looks sound". Explain why that is not enough, and what you would check instead.

    What it tests
    The faithfulness of shown reasoning, checking an answer through the facts and rules it depends on, and keeping people accountable for decisions about others.
    First step
    Separate the decision from the reasoning on screen. Research found reasoning models mentioned a hint that had decided their answer in only a quarter to two fifths of cases, so a sound-looking chain proves little. Check the inputs, each eligibility rule and the arithmetic yourself, against the lender's written policy.
    Follow-ups

    From the lesson Reasoning models: what 'thinking' changes

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

  8. Mid to senior

    Your team wants a reasoning mode switched on for every request in a customer support tool. How would you decide whether that is worth it, and what would you measure?

    What it tests
    Where thinking helps and where it does not, its cost in time and tokens, and a fair comparison on the team's own requests.
    First step
    Sort a sample of real requests by type. Simple questions and rewording gain little from thinking, while multi-step cases, such as a refund under several rules, can gain a lot; every thinking token adds time and counts against the allowance, so cost each type before deciding.
    Follow-ups

    From the lesson Reasoning models: what 'thinking' changes

    In the answer:

    • Model answer
    • Follow-up answers
    • Common mistakes
    • Rubric
    • Variants
    • about 3 minutes

All AI Literacy & Prompting interview questions · The AI Literacy & Prompting track