Almost every AI system that can hold a conversation traces back to a single paper published in 2017, with a title that reads like a boast: "Attention Is All You Need." Its eight authors, then at Google, were trying to improve machine translation, and their claim was narrow and technical, that the recurrent and convolutional networks which dominated language processing could be discarded entirely and replaced by a mechanism called attention. The narrow claim turned out to be a foundation. The architecture they introduced, the Transformer, is the T in GPT and the scaffolding under every large language model since.