This article returns to the release of ChatGPT to study the mechanics of Large Language Models (LLMs) and to counterpose them against established theories of language and writing. Its aim is to explore the hypothesis that ‘generated writing’ is “the beginning of new writing”, a writing that has its own cultural logic and tendencies. To show this, the article revisits philosophical theories of writing, in particular those of Jacques Derrida and André Leroi-Gourhan, and reads them in light of the computational methods of LLMs to understand in what way ‘generated writing’ diverges from the writing we have known before.