Hallucination
A hallucination is output from a language model that is fluent and plausible but factually incorrect or unsupported by any source. It arises because models predict the next token from learned statistical patterns rather than retrieving verified facts. Causes include gaps in training data, an outdated knowledge cutoff, ambiguous prompts, and the stochastic nature of sampling-based decoding, which can favor likely-sounding but false continuations.
Definition
A hallucination is output from a language model that is fluent and plausible but factually incorrect or unsupported by any source. It arises because models predict the next token from learned statistical patterns rather than retrieving verified facts. Causes include gaps in training data, an outdated knowledge cutoff, ambiguous prompts, and the stochastic nature of sampling-based decoding, which can favor likely-sounding but false continuations.
Examples
- ChatGPT inventing a non-existent court case citation in a legal brief.
- An assistant fabricating an API method name that the SDK does not expose.
- Grounding answers with retrieval-augmented generation to cite real source documents.
Related terms
No credit card required
Frequently asked
- What is Hallucination?
- A hallucination is output from a language model that is fluent and plausible but factually incorrect or unsupported by any source. It arises because models predict the next token from learned statistical patterns rather than retrieving verified facts. Causes include gaps in training data, an outdated knowledge cutoff, ambiguous prompts, and the stochastic nature of sampling-based decoding, which can favor likely-sounding but false continuations.
- How is Hallucination used in practice?
- Hallucination is commonly used in scenarios such as: ChatGPT inventing a non-existent court case citation in a legal brief.; An assistant fabricating an API method name that the SDK does not expose.; Grounding answers with retrieval-augmented generation to cite real source documents..
- How does yno.ai support Hallucination?
- yno.ai supports Hallucination as part of its agentic AI platform — combining multiple frontier models with extensible tool use.
- What other concepts are related to Hallucination?
- Hallucination relates to Retrieval-Augmented Generation, Knowledge Cutoff, Temperature (Sampling), Inference. Each links to its own glossary page below.