How AI Learns Anything
AI is never programmed with facts or rules. It learns by making predictions, measuring how wrong they were, and adjusting billions of tiny settings to be less wrong next time. Repeated enough, mistakes become knowledge.
Nobody typed it in
Nobody gave the AI a list of facts. So how does it know this one?
No one wrote “Eiffel Tower → Paris” into it. It picked this up the way it picks up everything.
Hide the next word
Training is one game, played over and over: hide the next word in real text, then guess it.
The text already holds the answer, so nobody has to mark the guesses.
An early guess
Early in training, its guesses are poor. Where do you think “Paris” ranked?
Measure the miss
The hidden word was “Paris”. It measures how far its score was from certain.
“Paris” got 31% when it should have been close to 100%. That gap is the mistake it learns from.
Nudge the settings
It nudges its billions of tiny settings so “Paris” scores a bit higher next time.
Again and again
Then the next sentence, and the next, across a huge amount of text.
Each guess nudges the settings a little. Repeated enough, the misses add up to patterns it gets right.
Guess, measure, nudge
AI learns by guessing hidden words, measuring the miss, and nudging its settings.
The settings are called weights. The measured miss is called the loss.