Hallucination
Concept
Training & Inference
Confident fabrication — the failure mode that shapes deployment.
You're looking at a concept explainer: what this term means in plain English, and the records it ties together.
Models sometimes produce fluent falsehoods, because next-token prediction optimizes plausibility, not truth. Mitigations — RAG grounding, citations, verifier models, calibrated refusals — reduce but do not eliminate it. It is the single biggest brake on trusting models in medicine, law and finance, and the reason this site labels every figure with its source.
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Sources
OpenAI on why models hallucinate ↗
Record reviewed as of 2026-01-15 — see Methodology.