Lesson 1 of 6
Where the knowledge comes from
9 min read
Ask an AI who wrote Hamlet, or how to boil an egg, and it just knows. But nobody sat down and typed those answers into it. So where did all of that come from?
Nobody typed in the answers
Picture a student who spent years in the biggest library on Earth — reading books, articles, forum threads, recipes, code, and endless back-and-forth conversation. They didn't memorise it word for word. But after enough reading, they picked up how ideas fit together: which words tend to follow which, how a recipe is shaped, how people answer a question. An AI learns the same way — by reading a staggering amount of human text and absorbing the patterns in it. That reading phase is called training.
An AI's knowledge isn't typed in. It comes from reading — a huge sweep of human writing during pretraining.
It's a reflection of what it read
Because everything it knows came from that pile of text, the pile shapes what it can do. Feed it plenty of stories and it gets good at stories. Feed it code and it can write code. Take a whole kind of writing away, and a whole ability goes quiet — there was simply nothing there to learn from. The model is, in a real sense, a mirror of its reading.
This is also why AIs pick up our habits — the good and the bad. If a lot of the text leans a certain way, the model leans that way too. It learned from us, so it reflects us: our knowledge, our styles, and our blind spots.
What went in shapes what comes out. An AI is a mirror of the text it was trained on — including its gaps and its biases.
It stops reading at some point
The reading doesn't go on forever. At some point training ends and the model is frozen — it knows the world up to roughly that moment and no further. Ask about something that happened last week and it may have never seen it. That edge is the model's knowledge cutoff.
This is why an AI can confidently miss very recent events. Unless it can look something up live, its knowledge ends where its reading ended.
A trained model is a snapshot. Fresh facts that appeared after its cutoff simply aren't in there.
What to carry forward
- —An AI's knowledge comes from reading human text, not from facts typed in by hand.
- —What it read shapes what it can do — remove a kind of writing and a skill fades.
- —Its knowledge stops at a cutoff; very recent things may be missing.
Where does an AI's knowledge actually come from?
Continue in the app
Take the whole How AI Learns course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.