AI basics
What is generative AI? A simple guide with examples
Generative AI creates new text, images, code, audio and video from a prompt. Learn how it works, real examples, what it is good and bad at, and the risks to know.
Short answer
is AI that creates new content, such as text, images, music, code or video, in response to a . It learns patterns from huge collections of examples and then produces new material that follows those patterns. ChatGPT, Claude and Gemini generate text; image tools like Midjourney, DALL·E and Stable Diffusion generate pictures.
How generative AI differs from other AI
Much of the AI you meet every day makes a decision about something that already exists: is this email spam, which film will you like next, is that a cat in the photo. Generative AI does something different. You describe what you want, and it produces something that didn't exist before: a paragraph, a picture, a melody, a piece of code.
It's built with deep learning: very large neural networks trained on millions or billions of examples. The model doesn't store and paste those examples. It learns the patterns behind them, then uses those patterns to make new ones.
How text generators work
like ChatGPT, Claude and Gemini are large language models. They write by predicting the next token, a word or part of a word, again and again. Training on a vast amount of text, then tuning with human feedback, makes those predictions useful and conversational. We explain the whole process in how ChatGPT works.
How image generators work
Most image generators are diffusion models. During training, the model sees real images with more and more random noise added, and learns to remove a little of that noise at each step.
To make a new image, it starts from pure static and removes noise step by step, steered by your text prompt, until a picture appears. It's a bit like a sculptor finding a statue in a block of marble. See it happen in how AI makes pictures and, in more depth, the diffusion lesson.
Examples of generative AI
- Text: drafting emails, summarising reports, answering questions, translating, tutoring.
- Images: illustrations, product mock-ups, design ideas, photo editing such as filling in a background.
- Code: assistants that suggest and write code, explain errors and write tests.
- Audio: natural text-to-speech voices, music generation, voice cloning.
- Video: short clips generated from text or from a single image.
Many newer models are multimodal: one model that can read and produce several kinds of media, like describing a photo or answering a question about a chart. The lesson on multimodal models explains how.
What generative AI is good and bad at
It shines when there are many acceptable answers: first drafts, brainstorming, rewording, explaining a topic in a new way, or making variations of a design. It's fast and tireless.
It struggles when there is exactly one right answer you can't easily check. It can invent facts, quotes and sources that sound real, called hallucinations. It can also miss things a human expert would catch, and image models still sometimes get small details wrong.
Risks to know about
- Made-up facts. Fluent doesn't mean true. Check anything important.
- Deepfakes. Realistic fake images, audio and video can be used to mislead. Learn the warning signs in spotting AI fakes.
- Bias. Models learn from human data, so they can repeat its stereotypes.
- Privacy. Don't paste passwords, personal details or confidential work into tools you don't control.
- Copyright. Courts and lawmakers in several countries are still working out how copyright applies to and to AI-made work, so check the rules before using outputs commercially.
Key takeaways
- Most AI sorts or predicts; generative AI makes something new.
- Text generators predict the next token; image generators turn noise into a picture step by step.
- It's great for drafts, ideas and variations, and weak at guaranteed facts.
- Watch for made-up facts, deepfakes, bias and privacy leaks.
Learn it properly, free
- Meet the chatbots: LLMsBeginner · 6 min
- How AI makes picturesBeginner · 6 min
- Diffusion: art from noiseAdvanced · 6 min
- Spotting AI fakesBeginner · 5 min
- Seeing and hearing: multimodal modelsIntermediate · 6 min
Terms in this post
Questions
Is ChatGPT generative AI?
Yes. ChatGPT is a generative AI chatbot built on a large language model. It generates new text, and some versions can also create images.
What is the difference between AI and generative AI?
AI is the whole field of making computers do intelligent tasks. Generative AI is the part of AI that creates new content, like text or images, rather than only classifying or predicting.
Is generative AI the same as AGI?
No. AGI, artificial general intelligence, means AI that can do most intellectual tasks as well as a person. Today's generative AI is impressive at many tasks but still makes basic mistakes, and whether or when AGI will arrive is debated.
What are examples of generative AI tools?
Text: ChatGPT, Claude, Gemini and open models like Llama. Images: Midjourney, DALL·E and Stable Diffusion. There are also generative tools for code, voices, music and video.
Sources and further reading
- Hugging Face Diffusion Models Course · Hugging Face · Free course
- Denoising Diffusion Probabilistic Models · Ho, Jain and Abbeel, 2020
- Elements of AI · University of Helsinki