Lessons › AI tools for traders › Hallucinations: sure-sounding and wrong
World 8 · AI tools for traders · lesson 4 · level 2
Hallucinations: sure-sounding and wrong
AI can invent facts that sound real, so a confident answer is not the same as a correct one.
In one line
The AI quoted an exact profit figure and a source, and neither one existed. How do you catch that?
Explained simply
Imagine a classmate who never says 'I don't know' and instead makes up an answer that sounds smart. An AI can do the same: it may invent a number, a news story or even a report that looks real. That is called a hallucination, so a sure-sounding answer still needs checking.
The lesson
A hallucination is when an AI states something false as if it were true, such as a made-up earnings figure, an event that never happened or a source that does not exist. It happens because the model generates likely-sounding text rather than looking facts up in a checked database. Warning signs include very precise numbers with no source, events after the model's knowledge cutoff, and answers that change when you ask again. US government guidance on AI warns that people often believe false answers because they sound confident.
A worked example
Illustrative example: you ask an AI twice for a company's latest quarterly profit. The first answer is 412 million dollars, the second is 380 million, and the company's own filing says 400 million. The two AI answers differ by 412 - 380 = 32 million. The first is 412 - 400 = 12 million too high, an error of 12 x 100 / 400 = 3%, and the second is 400 - 380 = 20 million too low, an error of 20 x 100 / 400 = 5%. Neither answer matched the filing, and only the filing counts.
The same idea at four levels
- Beginner. A hallucination is when an AI makes something up and says it as if it were true.
- Foundation. Confident words and exact-looking numbers do not make an answer correct.
- Intermediate. Red flags: precise numbers with no source, events after the AI's knowledge cutoff, and answers that change when you ask twice.
- Advanced. Open every source an AI quotes, because it can invent realistic report titles, quotes and links, and check each figure against the company's own filing or the exchange's data.
- Expert. Hallucinations come from how the model works, predicting likely text rather than retrieving checked facts, so even a tool that is usually right needs the same checks every single time.
Mistakes to avoid
- Trusting an AI's number because it is very precise.
- Using a source an AI quotes without opening it to check that it exists.
- Asking about very recent events and assuming the AI knows them.
Check yourself
What is an AI hallucination?
A false statement said as if it were true. The AI states something made up with a straight face.
Does a confident tone prove an answer is right?
No, confident and correct are different things. Anything can be said confidently, true or not.
Which is a warning sign of a hallucination?
A very precise number with no source. Precise but unsourced numbers need checking first.
You ask the same question twice and get two different numbers. What does that tell you?
At least one answer is wrong, so check the source. Two different answers cannot both be correct.
An AI quotes a report and gives a link. What should you do first?
Open the link and check the report exists and says that. AIs can invent links and reports that look real.
Goal of this lesson: Recognize AI hallucinations such as made-up numbers, events and sources.