What Is Hallucination?
A hallucination is when an AI model produces information that looks fluent and convincing but is wrong or has no source.
How does it work?
Language models work by producing likely text, not through a mechanism that checks truth. So they can produce a fluent answer even on a subject they do not know. The risk grows with missing context, vague questions and topics absent from the training data.
Where is it used?
- Hallucination is not a use case but a risk. It is especially critical in areas such as legal, finance, healthcare and customer commitments.
Points to watch
- To reduce the risk, answers should be grounded in sources and cite them (RAG (Retrieval-Augmented Generation)).
- Instructions and tests should be designed so the model says when it does not know.
- Critical outputs should pass human approval.
- The risk cannot be eliminated; regular evaluation tests are needed.
What does Huaris AI do about it?
Huaris AI makes source citation, evaluation sets and human approval steps part of the solution, and states clearly that the risk cannot be eliminated.
Let's work together
Let us listen to your processes and goals, and evaluate together where AI can add value.
Send an email