Glossary
Hallucination
A model producing fluent, confident output that is not true, the failure mode that makes evaluation non-optional.
A language model predicts plausible continuations. When it lacks the fact, it produces something that reads exactly like a fact, with no change in tone to warn you.
The mitigations are grounding (retrieve the fact rather than recall it), citation (make the source checkable), and measurement (a labelled question set that turns the rate into a number).
Grounding reduces it but does not eliminate it, because a model can also misread a passage it was correctly given. That is why citation matters as much as retrieval: a reader who can check the source in one click can catch the residual errors that measurement alone will not.
Commonly misunderstood: Teams frequently report that a model 'mostly gets it right' without having measured. The measured rate is usually higher than the impression.
Related terms, in context
The concepts you almost always meet alongside hallucination.
- RAG
- Retrieving relevant passages from your own documents and giving them to the model, so answers are grounded and citable.
- Evaluation
- A labelled test set that turns 'it seems good' into a number you can regress against.
- Red teaming
- Deliberately attacking your own AI system to find failures before users or attackers do.
Where this shows up in our work
Hallucination is not an abstraction for us. It is a decision we make on live projects. It shows up most directly in ai evaluation & red teaming, rag & knowledge retrieval, where getting it wrong has a cost someone can measure.
If you are evaluating a vendor on this, the useful question is not whether they can define the term. It is what they measure, what they would refuse to do, and what happens in their system when the assumption behind hallucination stops holding.
Questions
What is Hallucination?
A model producing fluent, confident output that is not true, the failure mode that makes evaluation non-optional.
What do people get wrong about hallucination?
Teams frequently report that a model 'mostly gets it right' without having measured. The measured rate is usually higher than the impression.
Does Orqent Labs build this?
Yes, AI Evaluation & Red Teaming and RAG & Knowledge Retrieval Systems. We work across India, covering all 19,238 PIN codes remotely.
Building something that involves hallucination?
We will tell you honestly whether it is the right approach for your problem.
Or email bd@dtrasglobal.com · call +91 74118 77878