Zone 01 · Q2

How does AI learn versus how it answers?

Two timelines, one insight. Two minutes.

Training and inference are two different jobs. Training is slow and expensive. Inference is fast and cheap.

Training. The long road.

1. Read 10 trillion words

2. Guess the next word

3. Compare guess to the real word

4. Nudge every weight a tiny bit

5. Repeat 10 trillion times

6. Freeze the weights. Call it a model.

Answering. The short road.

1. Read your question

2. Run it through the frozen weights

3. Pick the next word

4. Add it to the answer

5. Stop when the answer is done

Training happens once. It takes months and burns a city block's worth of electricity.

Answering happens every time you ask. It takes milliseconds. The model does not learn from your question. It only uses what it learned before.

This is why the same model gives different answers on different days only if the prompt changes, the sampling changes, or someone retrains it. The underlying weights sit still.