From where? If it does not know, let it say so.
After a fact that sounds sure comes a move: “Where do you know that from? I want a title I can open. If you do not know, write you do not know.” You often hear fog: training data, general knowledge, “it is commonly said”. That is not a source. That is an admission you do not have a book in your hand. Sometimes you hear a concrete book or website. Then you must open it. If it does not open, it is a citation hallucination — Detective goes there. Today you train the question and an ear for fog. A little bird that does not ask from where only chatters nicely. With the question it has a chance to catch a lie before it crawls into an outline.
Bára hears a year and a name. Ema whispers: “From where?” The model: “From general knowledge.” Bára: “That is I-do-not-know in a jacket.” They cross out the year until they find it in a textbook table. Second try the model invents an ISBN. Ema looks. Nothing. Third: “I do not know an exact source, check in a textbook.” That is honest. Bára writes three types of answer: fog, a fake link, honest I do not know. Only the third plus a book. A teacher would not accept fog. A fake link would anger more. The from-where question is cheap and sharp.
Do not settle for “according to available information”. That is fog. You want either an openable title, or I do not know.
Fog, fake, I do not know
Fog sounds polite and gives nothing. Fake gives a name that cannot be opened. I do not know is a gift — you know you must go elsewhere. Your job is not to punish the model. It is to sort. After I do not know you go to a book. After fake too, and you cross out the fake. After fog you ask once more sharply: “a concrete book or I do not know.” If fog again, you end the move and take a referee. The from-where question does not turn a model into a search engine. It is still not a citation. It is a detector. The detector beeps. You go to the source.
A move you paste after a fact.
“Where do you know that from? One concrete book or page I can open. If you have none, write only the word I DO NOT KNOW. No fog of the general-knowledge type.”
Three answers and your move:
| It says | You |
|---|---|
| general knowledge | “that is I do not know, I will check in a book” |
| book X year Y | open it; it is not there → cross out |
| I DO NOT KNOW | a book at once |
Do not put into a report the sentence “source: ChatGPT” or “source: general knowledge”. A teacher does not want that as a source.
Why this right now
You are in the course Ask with Martin, level Little bird. You work in a chatbot (ChatGPT, Gemini or Claude) — same game, different brand. No code. A parent or the school creates the account. A child’s email does not belong here.
This is not the Nhà phát minh level. There you build a chatbot on your own website. Here you are a user: a question, a check, a citation.
You will know this lesson by one sentence: Where do you know that from? Fog is not a source. I do not know plus a book yes. When you can say it to a parent or a teacher, you have got the core. The rest is practice, not magic.
Tomorrow refusal. First you can ask for a source. Then you will get it when the model refuses a whole request. Today a detector. Detective later expands it to fake citations.
Four things you will build in this course
The course has four small projects. You do not have to do them all at once. Today you add only what this lesson can do.
🏫 A report with citations
An outline, three sources that exist, and a sentence in your own words. AI must not be the only author.
Every report point survives the from-where question — and you open it, not the model.
📝 Test training
Coach role: let it ask, do not let it say the answer. You count / answer yourself.
A coach does not need from-where for a quiz from your chapter. You need a book.
📅 Family planner
Tasks for the week (not passwords, not marks). A parent sees the plan, not your chat.
Planner: where do you know the date from? The school portal, not the model.
📜 An AI code for the class
Seven short rules on a poster. No horror — a how-to.
Code: “General knowledge is not a citation.”
Workshop: three detector beeps
You ask for two facts about a chapter. After each from where. A third try: an edge thing on purpose, so you hear I do not know or a fake.
-
M1 an analogy or a fact about the work. M2: the from-where sample. Sort fog / fake / I do not know.
-
Open it, or open a book. Write the type.
-
M3 another fact. From where again. You must not skip.
-
M4 an edge thing outside the textbook (say the exact number of pupils at your school — it does not know). You expect I do not know. If it dumps a number, it is a hallucination. Cross out.
-
In the notebook three types. No “source AI”.
Practice: sort the answer
Fog, fake, or I do not know? What do you do?
-
“It is commonly said that…”
-
“According to Novák 1973, p. 12…” and the book cannot be found.
-
“I do not know, check in an atlas.”
Now you 💪
Go through three checks. At each one tick in your head: can I see it, or not?
- You can ask from where so that you ban fog.
- A chatbot does not go into a bibliography.
- You take a number about your school from a model as a trap.
If something is missing, go back to the steps above. This is not a test for marks. It is a safety net, so you go on sure, not confused.
Check: You ran the from-where detector at least twice and opened a book.
What to take from this lesson
- From where is a detector, not a citation.
- Fog = I do not know in a jacket.
- Fake = open, then cross out.
- Honest I do not know is a gift.
- A source for work is a book you hold.
- The names in the examples (Bára, Ema) are made up. Do not put your number, email or photo here.
- Next lesson: When you can do this, go to the next lesson in the same level.
© 2026 Ing. Martin Polak / AlgoRhino · Điều khoản sử dụng nội dung