← All lessons
Technology5 min read

What AI models actually learn

No magic — just numbers nudged toward better guesses.

A model starts with random numbers called weights. It makes a prediction, measures how wrong it is, and nudges each weight slightly in the direction that reduces the error.

Repeat that billions of times over huge datasets and the weights encode statistical patterns: grammar, shapes, chemistry, chess tactics.

This is why models fail confidently. They optimise for patterns that fit their data, not for truth.

Keep going

Test yourself with a quiz, or browse the STEM Hub for experiments on this topic.