Understanding Hallucinations

Building on what you learned in Module 1.

What you'll leave with

By the end of this lesson, you'll be able to identify which categories of AI output carry the highest risk of hallucination — and adjust your trust accordingly.

Why this matters

Module 1 introduced the concept of hallucination — AI generating false information with the same fluency and confidence as accurate information. This lesson takes that understanding and makes it practical: which types of output are most at risk, and what does that mean for how you use the tool?

The idea

A useful mental rule: the more specific the claim, the more careful you should be. "AI is a general-purpose technology" requires no verification. "This drug was approved by the FDA on March 14, 2022" absolutely does.

This is not a reason to avoid AI. It is a reason to use it with the right category in the right way.

The teaching block

High-risk content types:

• Specific citations: books, papers, articles, studies — AI will invent plausible-sounding titles and authors
• Statistics and data: precise numbers, percentages, and figures are frequently fabricated
• Dates and timelines: especially for events outside major historical moments
• Biographical details: facts about real people, especially those who are not widely covered
• Recent events: anything after the training data cutoff
• Niche or specialised information: the less common the topic, the higher the risk
• Legal and medical specifics: particular laws, drug interactions, dosages, procedures

Low-risk uses (where hallucination is less consequential):

• Drafting and writing: AI is generating language, not facts
• Explaining well-established concepts: things that are stable, widely documented, and not time-sensitive
• Brainstorming: the goal is option-generation, not factual accuracy
• Summarising content you provided: AI is working from your text, not generating facts independently
• Formatting and restructuring: turning bullet points into prose, or prose into an outline

Example

Ask AI to give you three statistics about a topic you know something about. Look up one of the statistics. Is it accurate? Is it even real?

This exercise is uncomfortable the first time. That discomfort is useful — it calibrates your trust in a way that reading about hallucinations never fully does.

Try this now

Ask AI to give you three statistics about a topic you know something about. Look up one of the statistics. Is it accurate? Is it even real?

This exercise is uncomfortable the first time. That discomfort is useful — it calibrates your trust in a way that reading about hallucinations never fully does.

Save this

Quiet takeaway

Knowing which outputs carry risk is not a reason to use AI less. It is a reason to use it more intelligently — with attention directed at the right things.

Next

Now you know which outputs carry risk. In Lesson 5.2, you learn to recognise the warning signs before you even verify.