Truth, Hallucination, and Misinformation

Picture a friend who is supremely confident, speaks in complete, polished sentences, and is also wrong about 10% of the things they tell you — but never once says "I'm not sure." That's roughly what it's like to talk to a large language model. It's not lying, exactly, because lying requires knowing the truth and saying something else on purpose. What it's doing is closer to confident guessing dressed up as fact. AI researchers call this hallucination.

Here's why it happens. A language model doesn't have a database of facts it looks things up in. It has learned, from enormous amounts of text, what words tend to follow other words. When you ask it a question, it's not retrieving an answer — it's predicting the most plausible-sounding continuation, one piece at a time, like an extremely well-read improv performer who has to keep talking no matter what. Most of the time that prediction lines up with the truth, because the truth is usually the most common pattern in its training data. But sometimes the most plausible-sounding answer and the correct answer part ways, and the model has no internal alarm bell to tell the difference.

This matters more as these tools get better at sounding right. A clumsy answer is easy to doubt. A fluent, well-structured, confidently-worded paragraph with a fake citation in it is much easier to believe — and that's exactly the kind of thing a hallucinating model produces. The danger isn't that AI gets things wrong; humans get things wrong all the time. The danger is that AI's wrong answers are dressed in the same confident voice as its right ones.

The fix isn't to stop using these tools — it's to change your relationship with their answers. Treat an AI's claim the way you'd treat a tip from a stranger: useful as a starting point, not as a citation. If a fact matters — a date, a statistic, a quote, a medical or legal claim — check it against an independent source before you repeat it. That single habit, applied consistently, neutralizes most of the risk.

This isn't unique to text, either. The same underlying issue — a system generating plausible content with no built-in fact-checker — shows up in AI-generated images, voices, and video (deepfakes). The skill you're building today, verifying before believing, is the same skill that protects you across all of it.