hallucination

When an AI states something false as if it were true, fluently and confidently, because it predicts plausible text rather than checking facts.

What it is

A large language model generates the most plausible next words from patterns, optimizing for sounding right, not being right. It has no built-in "I don't know." So when it has the answer and when it's guessing, the output looks identical: clean prose, specific details, real-looking citations, no hedging. A hallucination is that second case. The model invents a fact, a quote, a number, or a citation and presents it with the same confidence it uses for things it actually got right. The bluff isn't honestly vague; it's overconfident and specific, which is exactly why it fools people.

Why CEOs care

The confidence is the danger, not the error rate. A person who isn't sure usually shows it. The model doesn't, so a wrong answer ships looking exactly like a right one. Lawyers keep getting sanctioned for filing briefs full of court cases that don't exist, because the fake citations looked real. The same failure lands in board decks, memos, and customer answers. The rates are better than they were and still nowhere near zero, especially on hard real-world documents in law, medicine, and finance. The CEO's job isn't to wait for a perfect model. It's to know where you can't trust it blind and build a verification step into the workflow before output flows into a real decision.

Where you'll see it

• A fabricated legal citation or statistic that looks real, complete with names and page numbers.
• A confident, specific answer about your business that's flat wrong because the model never had the facts.
• A polished research summary that invents a source or misquotes a real one.

Full read

For the CEO-length version, see Why AI sounds confident when it's wrong.

Related

• What is a large language model?
• Why AI sounds confident when it's wrong
• What is a context window?