M02-U01 · 6-8 minutes

Generative AI Can Create More Than Text

Work through the explanation, apply it to the example, and complete the quick check before continuing.

What does generative mean?

Generative AI creates content from an input and the patterns learned during training.

The new content may be text, an image, audio, video, code, or data in a set format.

The word generative tells us what kind of result the system can produce. It does not tell us everything about how the model works.

A text model may write an answer, summary, translation, email, plan, or story.

An image model may create a new picture, change part of an existing image, or make a new version in another style.

An audio model may create speech, music, or sound effects. It may also change one voice into another.

A video model may create a short clip, add movement to an image, or change part of existing video.

A coding model may write a function, query, test, or setup file. Code still needs testing and review.

A model may return a table, list, or data in a format requested by another system.

A system can work with more than one kind of content. This is often called multimodal.

For example:

The input is an image and text. The output is text.

Another system may receive text and create an image. The input and output types are different.

Suppose we are planning a public tree-planting event.

System	Input	Generated result	What still needs checking
Text assistant	Event details	Short announcement	Dates, place, contact details
Image generator	Visual description	Poster image	Text inside image, suitability, rights
Speech generator	Approved announcement	Audio message	Voice permission and clear speech
Code generator	Form needs	Registration form code	Security, errors, and correct behaviour

All four systems generate content. Their outputs and risks are different.

Imagine a photo library.

You search for red flowers. The library returns three photos that already existed. That is retrieval.

Now imagine a model creates a new picture from the words red flowers beside a blue gate. That is generation.

A product may use both. It may find reference material and then create a new result.

In the next units, we will study a common way that text models generate language. They work with tokens and predict the next token again and again.

This is a useful way to understand text generation.

Do not turn it into a rule for every image, audio, or video model. Other kinds of models may use different methods.

A service finds and returns an existing training video from its library. What is proven?

A. Video generation. B. Video retrieval. C. Deep learning. D. An AI agent.

Check the answer

Answer: B. The service returns an existing item. There is no proof that it creates a new video.

Generative AI can create many kinds of content.
Input and output may use different media.
Generative describes a result, not one model design.
Retrieval finds existing material. Generation creates content.
We will use text models as our main example, not as an explanation of every model.

Next, we will see how a text model breaks language into tokens.