Meaning
A phenomenon where an AI model generates factually incorrect, nonsensical, or fabricated information with high confidence.
Three examples from work
Quality testing
QAThe AI bot invented a non-existent API parameter in its generated code snippet.
DevThat is a classic hallucination; we must inject the OpenAPI schema into the context.
Compliance audit
LegalCan the medical chatbot hallucinate drug dosage recommendations?
AI LeadWe added strict deterministic verification and fact-checking filters to prevent that.
Customer review
ClientThe assistant claimed our office is in London, which is incorrect.
POWe will tighten grounding prompts and require citations for all factual claims.
Mitigate hallucinations through grounding, search retrieval (RAG), few-shot guardrails, and self-consistency validation.