Reliability, Hallucinations and Verification

Why this lesson matters

AI systems can produce statements that sound convincing but are unsupported or false. This is often called hallucination. Reliability improves when users separate brainstorming from factual claims, request sources, verify primary evidence and keep humans responsible for consequential decisions.

Learning objectives

Recognise confident errors and apply a practical verification workflow.
Connect the concept to a real-world example.
Identify one limitation or responsible-use consideration.

Input
data, instruction or signal

→

System
rules, model or process

Key ideas

Confidence of tone is not evidence.
Verification should match the stakes.
Trusted documents and retrieval can improve grounding but do not eliminate errors.

Real-world lens

When evaluating this technology, ask what problem it solves, what information it depends on, how success is measured and what happens when it is wrong. This habit is more durable than memorising product names.

Hands-on activity

Create a three-level verification checklist for low-, medium- and high-stakes AI outputs.

Knowledge check

Which statement best reflects responsible technology learning?

A polished output should be trusted without checking. Understand the system, test the result and match verification to the stakes. New technology removes the need for human judgement.

Lesson summary

Confidence of tone is not evidence. Verification should match the stakes. Trusted documents and retrieval can improve grounding but do not eliminate errors.

← Prompting Well Responsible AI →