Quick Answer
Generative AI is software that creates new content: text, images, audio, video, code, and more. It works by learning patterns from enormous training datasets, then using those patterns to produce original output in response to a prompt. ChatGPT, Claude, Midjourney, and GitHub Copilot are all generative AI. It is different from traditional AI, which classifies or predicts rather than creates.
Glass sphere containing swirling purple and green plasma energy, suspended in dark space: a generative AI model contains compressed patterns from vast training data, releasing new content on demand.
A generative AI model is like a sphere that has absorbed vast amounts of human-created content during training: when you prompt it, it releases that compressed knowledge as new, original output.

Traditional AI vs generative AI

Traditional AI software classifies, predicts, or detects. It answers yes-or-no or multiple-choice questions:

  • Is this email spam? (yes or no)
  • Which product will this customer buy next? (pick one from a list)
  • Does this X-ray show a tumour? (probability score)

Generative AI produces open-ended new content:

  • Write a business proposal for a fintech startup in Vienna
  • Create a photorealistic image of a mountain landscape at sunset
  • Refactor this Python function to use async/await

The shift is significant. Previous AI required you to define the output categories in advance. Generative AI accepts almost any instruction and creates something new.

The four main types of generative AI

Text (LLMs)
ChatGPT Claude Gemini Mistral Generate text, summarise documents, write code, answer questions
Images
Midjourney DALL-E 3 Stable Diffusion Flux Generate images from text descriptions, edit photos, create product visuals
Audio and voice
ElevenLabs Suno (music) OpenAI TTS Convert text to natural speech, clone voices, generate music
Video and code
Sora (OpenAI) GitHub Copilot Lovable Cursor Generate video from text, write and complete code automatically

How it works: the pattern compression idea

Training a generative AI model works like this:

  1. Collect training data: For a language model, this is text from the internet, books, and other sources. Hundreds of billions of words.
  2. Train the model: The model processes this data repeatedly and adjusts its internal parameters (billions of numbers called weights) until it can accurately predict patterns in the data. This takes weeks on thousands of specialised chips.
  3. Compress the patterns: The result is a model that has encoded the statistical relationships in the training data into its weights. It does not store the original text. It has absorbed the patterns.
  4. Generate: When you give it a prompt, the model uses these patterns to generate the most likely continuation, word by word (for text) or pixel by pixel (for images).

A useful analogy: imagine someone who has read every book ever written. They have not memorised the books, but they have deeply absorbed how language works, how arguments are structured, and what tends to follow what. When you ask them to write something, they produce new text informed by all of that absorbed knowledge.

What generative AI is not

It is not a search engine. It does not retrieve existing content from the internet. A language model generates responses from its training data, which has a cut-off date. It may have no knowledge of events after that date.

It is not always right. It generates plausible output, not verified facts. It can confidently state incorrect information.

It is not conscious. It has no understanding, intentions, or feelings. It processes inputs and generates outputs according to learned patterns.

It is not magic. The output quality depends directly on the quality and specificity of the prompt you provide.

Training Data collection Billions of examples: text, images, code, audio. The model sees this data once or multiple times.
→
Training Pattern learning The model adjusts billions of internal parameters to minimise prediction errors on the training data.
→
Deployment Prompt input A user provides a prompt: a text description, an image, a partial piece of code, or a question.
→
Deployment Content generation The model generates new content one token (word fragment) at a time, using learned patterns to determine what comes next.

The business significance

Generative AI matters for businesses because it automates creative and analytical work at a scale and speed that was previously impossible:

  • Writing: First drafts of reports, proposals, emails, and marketing copy in seconds
  • Code: Junior-level coding tasks completed by AI in minutes instead of hours
  • Images: Product photos, marketing visuals, and illustrations generated without a designer
  • Summarisation: 100-page documents reduced to a structured brief in under a minute
  • Translation and localisation: Content adapted for new markets instantly

The economic case is strongest for high-volume, repeatable tasks where speed and cost matter more than perfect originality.

What’s next

Further reading