a ranked list of every possible next token with a percentage chance attached to each. If the text so far is "The capital of France is," the model might assign: