What is Hallucination?

AI hallucination occurs when a large language model produces text that sounds plausible and authoritative but is factually wrong, made up, or unsupported by its training data or the context provided. The term comes from the way the model 'perceives' something that isn't there — generating a citation that doesn't exist, a statistic that was never published, or a description of an event that never happened.

Hallucination happens because LLMs are fundamentally text-prediction engines. They are optimised to produce coherent, contextually appropriate continuations of text — not to retrieve verified facts from a knowledge database. When the model doesn't 'know' something, it fills the gap with statistically plausible text rather than acknowledging uncertainty.

For production AI systems, hallucination is one of the primary engineering challenges. Mitigations include: grounding the model in real documents via RAG (Retrieval-Augmented Generation), using structured outputs to constrain what the model can say, adding validation layers that verify outputs against a ground-truth source, and designing human-in-the-loop checkpoints for high-stakes decisions.