Hallucination
When AI confidently says something that sounds right but is made up.
In everyday terms
Because an LLM generates likely-sounding text, it can invent quotes, statistics or references that don't exist, in a completely confident tone.
For professionals
Fluent but unfaithful output: unsupported by the input or by fact. Reduced (not eliminated) by retrieval, citations, better training and lower temperature.
Think of it like…
A student who didn't study but writes a fluent, confident essay anyway.
You've already seen it
News stories about lawyers citing court cases that ChatGPT invented.
Myth vs reality
Myth: If the AI sounds confident, it's correct.
Reality: Confidence of tone has no link to accuracy. Verify important facts.
Quick check
Best defence against hallucinations on important facts?
- Ask it to be confident
- Check against reliable sources
- Raise the temperature
- Use more emojis
Show answer
Check against reliable sources: Verification is the only reliable guard.
Builds on
Related
Retrieval-Augmented Generation (RAG) · Temperature · Large Language Model (LLM)