Limits, mistakes and hallucinations.
AI can write a perfect-looking answer that is completely wrong. If you use AI for study, work or religion without knowing this, you can be badly misled. Today you'll learn why AI makes mistakes and how to check its work.
Hallucinations
A hallucination is when an AI confidently states something false: a fake fact, a made-up quote, a book that doesn't exist, a wrong date, or an invented hadith reference.
Why it happens: the model predicts likely-sounding text (Lesson 2). "Sounds right" and "is right" usually overlap, but not always. When it doesn't know, it may still produce something that sounds like an answer.
AI can invent hadith, wrong references or mix up rulings from different schools. For religious questions, check with reliable sources and scholars, and use tools that show their sources (our AI Scholar shows sources for this reason). Never treat an AI answer as a fatwa.
Other limits
| Limit | Example |
|---|---|
| Outdated knowledge | Doesn't know last month's news unless it can search |
| Maths and counting slips | Mistakes in long calculations or counting letters |
| Bias | Assumptions based on patterns in its training data |
| Overconfidence | Rarely says "I don't know" unless asked to |
| Missing context | Doesn't know your company's rules or your exact situation unless you tell it |
| Privacy risk | Anything you paste may be stored or used, depending on the tool's settings |
An AI gives you a perfect-sounding citation: "Journal of Indian Economics, 2019, page 45". What should you do?
AI can invent realistic citations. Always verify sources before using them.
The checking routine
- Ask for sources, then open and check them yourself.
- Cross-check important facts with a second reliable source.
- Ask the AI to double-check: "Are you sure? List anything you're uncertain about."
- Use your own knowledge: if something feels off, investigate.
- For numbers, recalculate or use a calculator/Excel.
Privacy: what not to paste
- Passwords, OTPs, bank or card details.
- Aadhaar, PAN or other ID numbers.
- Your company's confidential documents (unless your company provides an approved AI tool).
- Other people's private information.
Not good. Unchecked AI text may include false facts, and many colleges treat undisclosed AI use as cheating.
Good. AI for speed, human for accuracy.
Module 1 test
What is machine learning?
ML learns from examples.
What is a token?
Models read and write in tokens.
AI is generally best at…
Summarisation is a strength; values and empathy need people.
What is an AI hallucination?
Always verify important information.
Your task (15 min)
Ask an AI three questions you already know the answers to (about your area, your subject, or a famous event). Check every claim. Did it make any mistakes? Then ask it for sources on one answer, and verify them.
Lesson 4 complete!
You know what hallucinations are, why they happen, the main limits of AI, and a checking routine. You finished Module 1! Next: prompting that actually works.
Next: Prompting that actually works →