Modern generative AI rests on a surprisingly simple idea. Feed a model billions of sentences and play games with it—hide words and make it fill in the blanks, or ask whether one sentence logically follows another. Do this at scale and the model learns grammar, context, and a good deal of world knowledge.

In this video, you'll learn how that pre-training stage works, how instruction fine-tuning turns a general model into one that follows commands, and how those two steps produced the explosion of models—closed-source and open-source—that now compete on capability. You'll also see where text fits in the wider world of image and multimodal generation.