Lesson 5 of 28 · AI foundations
What is generative AI?
Generative AI is artificial intelligence that creates new content, such as text, images, audio, video or code, in response to a prompt. It learns patterns from very large amounts of training data and then produces new output that follows those patterns. Chat assistants, image generators and AI coding tools are everyday examples of generative AI.
Generative AI versus predictive AI
Many AI systems are predictive: they sort, score or label things that already exist. A bank's model flags a suspicious payment; an email filter labels spam. Generative AI does something different: it produces something new, such as the email reply, the product photo or the code.
Think of the difference between a film critic and a screenwriter. Predictive AI is the critic, judging what is in front of it. Generative AI is the writer, producing a new draft from everything it has absorbed.
How generative AI works, step by step
Every generative AI tool follows the same broad recipe, even though the details differ between text, image and audio models:
- Training: the model studies a very large collection of examples, such as web text, books, code or images paired with captions, and learns the patterns in them.
- Prompting: you describe what you want in a prompt, the instruction or question you type.
- Generating: a text model predicts the next token (a small piece of a word) again and again until the reply is complete. Many image models use diffusion: they start from random noise and remove it step by step until a picture matching the prompt appears. Earlier image generators often used generative adversarial networks (GANs), introduced in 2014.
- Variation: generation involves controlled randomness, so the same prompt can give a different result each time.
Types of generative AI
Generative AI is usually grouped by what it produces. Many newer models are multimodal, meaning they can take in or produce more than one type, such as reading a photo and answering in text.
- Text: chat assistants, writing tools, summarisers and translators, powered by large language models.
- Images: tools that create or edit pictures from a description.
- Audio: text-to-speech, voice cloning and music generation.
- Video: short clips generated from a text prompt or a still image.
- Code: assistants that write, explain and fix software.
What generative AI is not
It is not a search engine. Unless a tool connects the model to web search or your documents, it does not look anything up; it writes what is statistically likely. That is why it can produce a convincing but false statement, known as a hallucination.
It is not a copy machine either, but it is not free of its training data. Output is newly generated, yet it can resemble existing work, and commercial-use and copyright terms differ between vendors and countries. Treat output as a draft to review, especially before publishing.
Where you meet generative AI
OpenAI launched ChatGPT on 30 November 2022 and brought chat-based generative AI to everyday users. Generative features now appear in many tools: writing help inside workspaces such as Notion AI, marketing-copy tools such as Jasper, image generators such as Midjourney, and coding assistants inside developer tools.
When you compare generative AI products, you are comparing two things: the underlying model, which sets the ceiling on quality, and the product around it, which decides the workflow, controls, integrations and how your data is handled.
Example
See generation in action
Try this promptWrite three different one-sentence slogans for a neighbourhood bakery that opens at 6 a.m. Then, in two sentences, explain how you came up with them.
Run the same prompt twice and compare. You will get different slogans, because generative models sample from many likely continuations instead of retrieving one stored answer. Its explanation of how it worked is also generated text, not a real view inside the model.
Key points
- Generative AI creates new content instead of only classifying or scoring existing data.
- Text models generate one token at a time; many image models refine random noise into a picture.
- The same prompt can give different results each time.
- Output can be wrong or resemble training data, so review it before you publish.
- Chat assistants, image generators and coding tools are all generative AI.
Try it on Mindber
Related terms
Related topics
Frequently asked questions
What is the difference between AI and generative AI?
AI is the whole field of making computers perform tasks that need intelligence. Generative AI is one part of it: systems that create new content. A fraud-detection model that scores transactions is AI but not generative; a chat assistant that writes a reply is generative AI.
Is generative AI output original?
It is newly generated rather than copied from a database, but it is built from patterns in training data and can sometimes closely resemble existing work. Copyright and commercial-use rules differ by country and by vendor, so check the tool's terms before publishing generated text or images commercially.
Do I need technical skills to use generative AI?
No. Most tools work through plain-language prompts typed into a chat box or editor. Results improve when you give clear context, examples and the format you want, which is the core of prompt engineering. Developers can also reach the same models through APIs.
Sources
- Generative Adversarial Networks. Goodfellow et al., arXiv, June 2014
- Denoising Diffusion Probabilistic Models. Ho et al., arXiv, June 2020
- Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1. National Institute of Standards and Technology, July 2024
- Introducing ChatGPT. OpenAI, November 30, 2022
Last updated