What generative AI actually is

The mental model that makes everything else make sense.

What it is

Generative AI systems learn patterns from huge amounts of text, images or audio, then produce new content that follows those patterns. A chatbot isn't looking answers up; it's predicting what a good continuation looks like.

Why it matters

If you know it predicts rather than retrieves, you'll know when to trust it, when to check it, and why the same question can get different answers.

Key ideas

• A model is trained once on a fixed dataset, then used many times.
• It predicts the next piece of text (a token), one at a time.
• Fluency isn't accuracy: it's optimized to sound right.
• Tools add search, files and code on top of the model to ground it.

Try it: Catch the prediction in the act

1. Ask any assistant the same open-ended question three times in fresh chats.
2. Compare the answers. What stayed the same, and what changed?
3. Ask for a fact from last week with web search off, then on.
4. Write two sentences on what the difference tells you.