How It Works
Hallucination
When a model generates fluent, confident-sounding text that is factually wrong or entirely made up.
Origin
The term has older roots in computer vision (filling in plausible detail in blurry images); its application to language generation appears in machine-translation research by at least 2017, becoming the standard term for this LLM failure mode as generative models scaled up around 2020–2021, well before it entered public vocabulary via ChatGPT in 2022.