Select a prompt. The chart shows what the model thinks comes next — and how confident it is.

↑ input tokens    model predicts what comes next ↓
The model assigns different probabilities to candidate next tokens based on the input context.

The model does not "know" the answer — it calculates probabilities. The word it picks is drawn from this distribution. Different contexts shift the whole distribution, which is why the same model can write poetry, explain tax law, and debug code.