Hallucination
When AI confidently generates incorrect information
What is Hallucination?
When an AI model generates confident but factually incorrect, fabricated, or nonsensical information. LLMs don't "know" facts, they predict likely text sequences, which means they can sound authoritative while being completely wrong. For content creators and consultants, hallucinations are the biggest risk when using AI for research or drafting, always verify claims, especially around token prices, protocol specs, and regulatory details.
How it works
Hallucination: confident but false
Key takeaways
-
Confident output that's factually wrong
-
Hard to detect without verification
-
RAG and grounding reduce but don't eliminate it
Real-world example
You ask an LLM for a citation and it confidently provides a fake paper title, author, and DOI. It sounds real because the model learned the pattern of academic citations, but the paper never existed.
