AX-100 · Lesson 3 of 12 · Module 1: How Assistants Work · 2:59
Why it gets things wrong
Assistants guess when they do not know, and a guess reads exactly like a fact: research published in 2025 showed that training rewards a guess over “I don’t know”. Slips cluster on checkable details (names, numbers, dates, quotes and links), on anything after the cutoff, on rare topics and on sums typed as text. Give it the facts, let it say “not sure”, ask for its source text, and check every high-stakes line at a trusted source. Worked example: a list of cat-safe plants that includes the daylily, which is highly toxic to cats.

What you will learn
- Explain why assistants sound sure when wrong
- Name where mistakes are most likely
- Ask in ways that make guessing less likely
The one thing to remember
Sounding sure is not being right.
Common mistake: Trusting the tone of an answer
Assistants sound the same when they guess as when they know, so a calm, fluent line can be the dangerous one.
The fix: Let it say ‘not sure’, give it the facts, and check every high-stakes detail at a trusted source.
Key terms
- Hallucination
- A confident answer that is false or made up.
- Checkable detail
- A name, number, date, quote or link you can verify.
- High stakes
- When a mistake could hurt someone or cost real money.
Part of AI Fundamentals (AX-100)
For anyone at work new to AI: how assistants answer, how to ask well, how to check the answer and keep data safe.
12 lessons, practice quizzes, a final exam and the Sophyx Certified AI Fundamentals certificate. Regular price USD 49.99. Your first certification is free. Free with any paid Sophyx plan.
Start the course free