2017 paper "Attention Is All You Need" proposed Transformer

The paper's historical significance lies in how completely it broke from the dominant sequence-processing approach of its time. By discarding recurrence entirely, the Transformer made it practical to train on much larger datasets, a shift that later architectures built upon. This lineage traces from the original Transformer to BERT and GPT, showing how the same building blocks were specialized for different tasks.

The architecture's reach goes beyond text: it became the standard for natural language processing and was extended to vision tasks through ViT and to multimodal models like CLIP. For readers studying for the G検定, the paper is presented not just as history but as the conceptual foundation for the AI systems in use today.