// How an AI learns — guess, measure mistake, adjust, repeat billions of times
💡 The core idea: Training an AI is like teaching a child to ride a bike. The AI makes a guess, sees how wrong it was, adjusts, and tries again. After billions of such tries, it gets very good. The "adjustment" is done by a mathematical process called backpropagation.
Step 0 / 6
Ready — click "step" to begin
Training Loss ↓ (lower = better)
How does AI learn?
Training an AI is all about repetition. The AI reads a sentence, tries to predict the next word, checks how wrong it was, then nudges its internal numbers slightly in the right direction. After doing this billions of times across terabytes of text, it gets very good at predicting words — and that's essentially what "intelligence" is in an LLM.