When an LLM confidently generates false or fabricated information.
1 min readupdated 2026-06-22
/ quick answer
Hallucinations happen because LLMs are next-token predictors, not fact retrievers. They generate what looks plausible given the context, which can include invented citations, APIs, quotes or numbers. Mitigations: RAG, tool calling, evals and human-in-the-loop. When an LLM confidently generates false or fabricated information.
When an LLM confidently generates false or fabricated information. Hallucinations happen because LLMs are next-token predictors, not fact retrievers. They generate what looks plausible given the context, which can include invented citations, APIs, quotes or numbers. Mitigations: RAG, tool calling, evals and human-in-the-loop. In practice: Asking 'give me the DOI for a paper on X' can return a perfectly formatted DOI that points to nothing. RAG grounding on a trusted corpus reduces this dramatically. This dictionary node is part of the Onexial knowledge graph and links to related concepts, workflows and tools below.
Definition
Hallucinations happen because LLMs are next-token predictors, not fact retrievers. They generate what looks plausible given the context, which can include invented citations, APIs, quotes or numbers. Mitigations: RAG, tool calling, evals and human-in-the-loop.
Example
Asking 'give me the DOI for a paper on X' can return a perfectly formatted DOI that points to nothing. RAG grounding on a trusted corpus reduces this dramatically.
Hallucinations happen because LLMs are next-token predictors, not fact retrievers. They generate what looks plausible given the context, which can include invented citations, APIs, quotes or numbers. Mitigations: RAG, tool calling, evals and human-in-the-loop.
What is an example of Hallucination?
Asking 'give me the DOI for a paper on X' can return a perfectly formatted DOI that points to nothing. RAG grounding on a trusted corpus reduces this dramatically.
Why does Hallucination matter for AI and automation?
When an LLM confidently generates false or fabricated information. It connects to the workflows, prompts and tool stacks linked on this page, so you can move from definition to execution without leaving Onexial.