What Is an AI Hallucination?

Also called AI hallucination

Short Answer

A hallucination is when an AI model states something false or invented as if it were true, such as a fake fact, a wrong number or a source that does not exist. It happens because a language model generates the most likely-sounding text, which is not always the correct text.

  • Taught in lesson 2

Why It Matters

Hallucinations are the main reason AI output cannot be trusted blindly. Anyone using or building AI must know when they are likely, how to reduce them and how to design checks, especially in fields such as health, law and finance where a wrong answer has real consequences.

How It Works

Two famous cases: in February 2023, Google's Bard chatbot gave a wrong fact about the James Webb Space Telescope in a promotional demo, and Alphabet's market value fell by about 100 billion dollars that day. In 2023, a New York lawyer was sanctioned for filing a brief with court cases ChatGPT had invented.

Hallucinations cannot be fully removed, but they can be reduced: ground the model in trusted documents with retrieval, ask for citations, check outputs automatically and keep a human reviewer where errors are costly.

Hallucinations are more likely when a question is about rare facts, recent events, precise numbers or sources, and when the prompt pushes the model to answer instead of admitting uncertainty. They are reduced, not removed, by grounding answers in retrieved documents, asking for citations and checking them, constraining output formats, letting the model say "I do not know" and keeping a person in the loop for important decisions.

Hallucination in Practice

Everyday Example

Asked for books on a niche topic, an assistant lists a real author with a title that author never wrote.

Industry Example: Pharma

In pharma, a hallucinated dose, interaction or reference could harm patients, so medical and regulatory uses of AI require every claim to be traced back to a verified source.

Where You Meet It

  • Invented references or quotes in AI-written text.
  • Wrong numbers in AI summaries.
  • Evaluation reports measuring groundedness.
  • Company AI policies that require a human to check outputs.

Key Ideas Around It

Grounding
Making answers depend on given sources.
Citation
A link from a claim to its source, which can be checked.
Groundedness
How well an answer is supported by its sources.
Abstention
The model saying it does not know.

A Common Mix-Up

A hallucination is not a bug that a future update will simply remove. It comes from how language models generate text, so applications must be designed to catch it.

Check YourselfName two ways to reduce hallucinations in a document assistant.

Ground answers in retrieved documents with citations, and let the model say it does not know when sources do not cover the question.

Learn It on ai2hired

S0 · #2What AI Really Is: AI vs Machine Learning vs LLMs

This free lesson teaches Hallucination from zero, with the job ads that ask for it.

Questions People Ask

Why do AI models hallucinate?

They are trained to produce plausible text, not to check facts. When they lack the right information, they still produce a fluent answer.

How do you reduce hallucinations?

Give the model trusted sources with retrieval, require citations, evaluate answers against known correct ones and add human review for high-stakes outputs.

Learn AI From Zero to Hired

Free lessons on YouTube, each one built from what real AI job ads ask for.