Hallucination

When an LLM produces confident-sounding output that is factually wrong or invented.

Hallucinations come from missing context, ambiguous instructions, or training-data gaps. Mitigations: ground the model with relevant docs (RAG), ask for sources, narrow the scope, and verify before merging.

See it in the glossary →