How AI works: overview and map

In short: Modern AI is mathematics at its core: a model with many adjustable numbers (parameters) is tuned with examples until it recognises patterns and makes predictions. This series explains step by step how that works.

In more detail: “AI” is an umbrella term. Underneath lies machine learning (models learn from data instead of fixed rules), under that deep learning (deep neural networks), and under that large language models (LLMs) like those in chat assistants.

In detail

• Rules versus learning: classic programs follow rules written by humans. Learning systems derive the rules from examples.
• The ingredients: data, a model (architecture), a goal (loss function), a learning method (optimiser) and computing power.
• Two phases: training (parameters are set, expensive) and inference (the finished model is used, cheap per request).
• Order of the series: basics and classic learning → neural networks → architectures → language models from token to chat → training at scale → images and speech → use, safety and limits.
• Getting started: What is machine learning?, Data, features and labels, Neurons, layers and activation functions.
• Going deeper: How a language model generates text: step by step.
• Older articles: AI, Generative AI, Transformer architecture.

See also: How a language model generates text: step by step, What is machine learning?