What do you already know?

This is not an entrance test. Choose the best answer based on what you currently think. Your result only suggests how actively to use the early explanations and checks.

Your answers are private. This page checks answers in your browser. It does not submit or store your result.

1. Which statement is the strongest evidence that a system uses machine learning? It follows a fixed list of if/then rules written by a developer. Its behaviour was fitted from examples and can be evaluated on new examples. It has a chat box and uses friendly language. Its product page calls it intelligent.

2. A generative-AI response is fluent and confident. What can you conclude? The response was independently verified. The response is probably copied from one exact source. The response is well-formed, but its claims still require suitable review. The model understood the topic as a qualified professional would.

3. You need the exact total of 38 invoices. Which first tool is usually the better fit? A calculator or spreadsheet formula. A text-generating AI asked to estimate. An image generator. A chatbot asked to choose a plausible total.

4. Which input is suitable for practice in a public AI tool? A real employee grievance containing names and contact details. An unreleased client proposal. A synthetic meeting note created for the exercise. Your organisation's production password-reset instructions.

5. What is inference? Collecting every possible training example. Using a trained model to produce a prediction or output for an input. Proving that every model response is true. Writing the rules for a traditional spreadsheet.

6. A chatbot gives a citation. What is the safest next step before relying on it? Trust it because citations cannot be generated incorrectly. Ask the same chatbot whether it is certain. Open the source, confirm it exists, and check that it supports the claim. Remove the citation and keep the claim.

7. Which request is the clearest task specification? Make this better. Write something professional. Summarise the supplied policy for new employees in five bullets, preserve the stated exceptions, and flag anything that needs HR confirmation. Act as a genius and never make mistakes.

8. When is search or retrieval different from generation? Search retrieves items from an indexed source; generation produces an output from model processing and context. They are always the same operation. Generation automatically checks every live website. Search always creates new facts.

9. An AI-assisted decision could materially affect a person's employment. What is most important? Use the output quickly so the process feels efficient. Hide the use of AI to prevent questions. Apply authorised human review, relevant evidence, policy, safeguards, and a route to challenge or stop the decision. Ask the model to declare itself unbiased.

10. What does a larger prompt guarantee? A correct answer. A current answer. Nothing by itself; relevance, sources, constraints, tool capability, and review still matter. That the model will disclose all limitations.

11. Which statement about human oversight is strongest? A person glanced at the final screen. A named person had the authority, evidence, time, criteria, and ability to change or stop the result. The user clicked approve after the AI selected the answer. Someone could theoretically review it later.

12. You cannot independently verify a high-consequence AI answer. What should you do? Use it because the wording is detailed. Use it but remove any uncertainty language. Pause, seek an authoritative source or qualified reviewer, or choose a safer process. Generate the answer repeatedly and use the most common version.

Starting guidance

Continue to Module 1

Diagnostic guide

12 questions
About 8-10 minutes
No pass or fail
Retry whenever useful

Answer every question, then use the explanation below each item to review your thinking.

AI Foundations course home · Contact