Architecture is how a neural network is built: which layers it consists of and how data flows through them. Architecture is what defines what a model can do — write text, recognize images or generate video.
Almost all modern language models are built on the Transformer architecture. Understanding its principles explains both the strengths of neural networks and their limits: the context window, hallucinations, the cost of generation.
This section is written without formulas or heavy math — only what helps you use AI deliberately.