Hallucination
What Hallucination Means
In AI, hallucination refers to a response that sounds fluent and confident but is incorrect, unsupported, or fabricated. The model may produce a made-up citation, an inaccurate fact, a false summary, or a detail that was never present in the source material. The danger is that the output often sounds plausible even when it is wrong.
Why It Matters
Hallucination matters because it affects trust. AI can be helpful in research, writing, coding, and support workflows, but hallucinated output can introduce error if users accept it too quickly. The problem becomes especially serious in legal, financial, medical, enterprise, or technical contexts where accuracy matters more than fluency.
Why It Happens
Language models generate responses based on learned patterns and current context, not by guaranteeing factual truth. If the prompt is vague, the source material is missing, or the model is pushed beyond what it can ground reliably, it may still generate an answer rather than admit uncertainty. That is one reason hallucination remains an important evaluation topic in AI.
Where It Appears Most Often
Hallucinations often appear in citation generation, factual question answering, document summarization, code explanation, and structured claims about topics that require precise grounding. They can also appear when users ask the model to act as if it has seen information it never actually received.
How People Reduce It
Teams often reduce hallucination risk by using better prompting, retrieval systems, source grounding, human review, and task-specific evaluation. In some workflows, models are instructed to say they do not know rather than invent a likely answer. Good product design often matters as much as raw model strength here.
Best Practice
If an AI answer matters for a real decision, verify it against trusted sources instead of trusting fluency alone. Better AI use begins when users treat confident language as something to evaluate, not something to assume is automatically correct.
Understand AI risk and model behavior more clearly with AI Days — practical explainers, model comparisons, and daily AI updates.