Why AI Hallucinates
62%
Lesson 5 of 8

Why AI Hallucinates

A fluent answer can still contain invented facts because generation optimizes for a plausible continuation, not guaranteed truth.

Active reading
Why this matters

A fluent answer can still contain invented facts because generation optimizes for a plausible continuation, not guaranteed truth.

Plausibility

The answer can sound right.

Evidence

Truth needs supporting sources or data.

Uncertainty

The model may not reliably know when it lacks evidence.

Understand the working model

Hallucination risk rises when you ask for obscure facts, exact citations, current information or details that are absent from the supplied context. The safer pattern is to provide evidence and ask the model to stay inside it.

For important decisions, separate generation from verification: let AI draft the answer, then check the claims against primary sources, your own data or a retrieval system.

Example
Use only the supplied policy text. If a requested fact is not present, write “not found in the provided source” instead of guessing.
Practice

Apply the pattern above to one real task you already do. Keep the first version small enough that you can inspect every input and output.

Common mistake

Never ask the model to invent a citation or source to make an answer look complete.

Key takeaway

Treat fluent output as a draft. For factual work, require evidence and verify important claims.

Next lesson

Better Instructions

Continue to the next lesson.

Continue →