Purpose
Use this lesson to use generated text as a draft that needs judgment, not as a guaranteed fact source. The aim is to produce work that a colleague can review and act on, not merely text that sounds convincing.
Lesson Explanation
Scenario: A tool confidently says a competitor opened an office in Cairo.
Better instruction: List the claim, identify what source would verify it, and mark the claim unverified until that source is checked.
A language model predicts plausible next words. Plausible wording can still contain invented names, dates, citations, or calculations. Ask it to distinguish known inputs, assumptions, and items requiring verification. No external claim is treated as true until a reliable source confirms it.
Work in a short cycle: provide the relevant input, state the result and limits, inspect the draft against the acceptance criteria, then revise the instruction when a requirement is missing. Keep the final human decision with the person responsible for the work.
Practice Questions
-
An AI tool states with total confidence that a specific law was passed in 2019, with no source given. What does this lesson say you should do with this claim?
View Answer
Treat it as unverified until a reliable source confirms it – a language model predicts plausible wording, which can include invented facts stated confidently.
-
Why can a language model produce a confident, fluent statement that is factually wrong, according to this lesson?
View Answer
Because it predicts plausible next words, and plausible wording can still contain invented names, dates, citations, or calculations.
-
A prompt asks an AI to list a company’s exact quarterly revenue figures from memory, with no source document provided. What risk does this lesson warn about?
View Answer
The figures may be invented but stated with the same confident tone as verified facts, since the model predicts plausible-sounding output rather than checking a real source.
-
Rewrite this prompt to reduce the risk of invented facts: “Tell me the exact founding date of this small local business.”
View Answer
Something like: “If you know the founding date of this business from a specific, citable source, state it and the source. If you’re not certain, say so explicitly rather than guessing.”
-
What three categories does this lesson suggest asking an AI to distinguish in its output?
View Answer
Known inputs, assumptions, and items requiring verification.
-
A report generated by AI includes a specific statistic with no citation. Based on this lesson, what should happen before that statistic is used in a real business decision?
View Answer
It should be treated as unverified and checked against a reliable source before being treated as true.
-
An employee says, “The AI wouldn’t make up something that specific and detailed – it must be true.” Does this lesson support that reasoning?
View Answer
No – specificity and confidence in the wording do not indicate accuracy; plausible wording can still contain invented details.
-
A prompt is written to explicitly ask the AI to mark any claim it cannot verify as “unconfirmed.” What principle does this reflect?
View Answer
Distinguishing known inputs, assumptions, and items requiring verification, rather than presenting everything with equal confidence.
-
Why might a language model include a citation that looks legitimate but doesn’t actually exist?
View Answer
Because it predicts plausible-looking text, including citation formats, without necessarily verifying that the specific source is real.
-
A team uses AI output containing a specific customer quote in a public report without checking whether that quote is real. What does this lesson suggest they should have done first?
View Answer
Verified the claim against a reliable source before treating it as true, rather than accepting confident-sounding output at face value.
-
What should never be treated as true, according to this lesson, until a reliable source confirms it?
View Answer
Any external claim generated by the AI.
-
A colleague asks an AI for “the most likely reasons” behind a trend, and the AI returns a confident, single explanation with no uncertainty noted. How might a prompt be improved based on this lesson’s principle?
View Answer
By explicitly asking the AI to separate confirmed facts from plausible hypotheses, rather than presenting one explanation as certain.