“Natively, a Large Language Model (LLM) primarily predicts the next token.”