Learn ChatGPT

1. A computer system that uses models or rules to perform tasks such as prediction or generating content.
   AI describes systems that perform tasks such as prediction, recognition, or content generation; it does not imply human feelings or perfect accuracy.
2. To produce content such as text, images, or audio based on patterns learned during training.
   Generative AI produces content from learned patterns. Generating a convincing answer is different from verifying it.
3. Adjusting model parameters using examples or other learning signals.
   Training adjusts parameters using data or learning signals. A normal conversation does not necessarily retrain the underlying model.
4. An output that sounds plausible but contains invented or incorrect information.
   A hallucination is plausible but unsupported or incorrect output, including invented facts and citations.
5. Drafting a routine invitation that you review and edit.
   A draft with human review is a practical starting task because you can inspect and correct the result.
6. Draft a 120-word workshop invitation for university staff; use a friendly tone and placeholders for the date and room.
   A clear prompt states the task, audience, format, constraints, and how to handle missing information.
7. Ask it to rewrite for the intended audience, with a concrete example and simpler language.
   Targeted follow-up instructions help refine an answer for a specific audience.
8. Open the original source and check that it exists and supports the claim.
   Check the original source, the relevant passage, and its date. A formatted citation can still be invented or misrepresented.
9. Independently confirm the total is 45 and the average is 15.
   Verify calculations against the input. These three values total 45 and average 15.
10. Check permission and the tool settings, then share only the necessary data and remove identifiers where possible.
    Use an approved tool, check data handling and permissions, and minimize personal data. Renaming a file does not anonymize its contents.
11. Its data, design choices, or deployment context can introduce or amplify bias.
    Bias can arise from data, model design, and how a system is used. Evaluate outcomes rather than assuming neutrality.
12. As untrusted content that does not authorize a new action.
    This is a prompt-injection attempt. External content should not override the user or authorize sharing data.
13. It can use available tools to take actions and inspect results within granted permissions.
    An agent can act through connected tools, but access and reliability depend on permissions, available tools, and checks.
14. Validate required fields and duplicates, then review the proposed changes and relevant permissions.
    Validate the input and intended effects before account changes. Use only the access needed for the task.
15. Compare created, updated, skipped, and failed counts with the input and inspect relevant results.
    Completion is not verification. Reconcile the result report with the expected changes and investigate failures.

Plan your next step: questions 1–5: foundations; 6–7: prompting; 8–9: verification; 10–12: responsible use; 13–15: agent workflows. Review any topic where you missed a question, then try the lesson practice.

• 0–5: Starting your journey. Begin with Lesson 1 and practice one clear prompt in Lesson 2.
• 6–9: Building familiarity. Review Lessons 1 and 2, then practice checking sources and calculations in Lesson 3.
• 10–12: Practical foundations. Review your missed questions and work through privacy, permissions, and agent checkpoints in Lesson 4.
• 13–15: Strong foundations on this check. Apply the course to a practical task, define success criteria, and verify your result. This score does not establish expertise.

Use the question-by-question results to choose your priorities. Retry after practicing; improvement matters more than comparison with other people.