Learn AI fundamentalsModule 3 of 4 · 0 done
Checking AI output
Spot hallucinations, verify claims, and set a review level that matches the risk of the task.
By the end you’ll be able to
- Recognise common hallucination patterns
- Verify claims with primary sources
- Match review effort to risk
Where hallucinations hide
Specifics are the danger zone: names, numbers, dates, quotes, citations, URLs and product features. Fluent hedging-free text about niche topics deserves extra suspicion. So do summaries that add details the source didn’t contain.
- Check every number and name against a source
- Click every link and citation
- Compare summaries with the original
Ground and verify
Ask the model to answer only from provided documents and to quote the passage it used. Then check those quotes exist. For facts, go to primary sources: official docs, the original study, the company’s own pages.
Review in proportion to risk
An internal brainstorm needs a glance. A customer email needs a careful read. Legal, medical, financial or public statements need expert review — AI can draft, but a qualified person signs off.
Fact-check an AI answer
Ask an AI assistant a factual question in your field that has a checkable answer.
Deliverable: An annotated answer and a personal verification checklist. Add it to your skills passport as a project when it’s done.
Check your understanding
Answer 3 scenario questions. Get 2 or more right to complete the module.
See how your evaluating output scores
A free test adds Evaluating output to your skill report and shows how close you are to the 80% job-ready target.