Large language models often feel like they are “thinking” or composing responses the way humans do. In reality, the mechanism behind text generation is much simpler and more mechanical.

At its core, an LLM generates text by repeatedly predicting the next token in a sequence.

Even though the underlying task is simply next-token prediction, the result can look surprisingly intelligent. Because the model has learned patterns from vast amounts of text, it can produce coherent explanations, write code, answer questions, and carry on conversations.

In this chapter, we will explore how this generation process works in practice.