DictionaryEvaluation
Hallucination
When an AI confidently makes something up, like a fake fact, a library that does not exist, or a wrong quote.
Definition
When a model generates text that is factually incorrect or inconsistent with the given context, often stated confidently, such as a nonexistent API, file, or citation. Anthropic's guidance to reduce it: let the model say it does not know, ground answers in direct quotes from provided documents, require a citation for each claim, and compare several runs (best-of-N). These techniques reduce hallucinations but do not eliminate them.
Example
The model suggests import { retryFetch } from 'axios', a function axios does not export. Running the code or checking the docs catches it.