Next-Token Prediction: The Core Mechanism

LLMs generate text by producing a probability distribution over possible next tokens and picking one, repeating this loop token by token.