The capital of India is…

This is the core generation loop. Given the current tokens, the model scores possible next tokens, turns scores into probabilities, selects one, appends it, and repeats.

Current tokens
 ↓
Model
 ↓
Scores for possible next tokens
 ↓
Probability distribution
 ↓
Select next token
 ↓
Append token
 ↓
Repeat

Click through the simulation. Notice how “New” then “Delhi” can appear as the sequence grows — each step only chooses the next piece.