Sorting What Exists vs. Making Something New | AI Learning Guide

Think about a spam filter. It reads an email and answers with one of two fixed words. The answer already existed; the model only had to pick it. Now think about asking a tool to write a short poem about rain. There is no menu of poems to choose from. The words have to be assembled into something that was not there before. That is the whole difference the word generative is pointing at, and it is a difference in the task, not in how clever the system seems.

0:00 / 0:00

Auto-play

Predictive models choose among answers that already exist. Generative models create an answer that did not exist before. That single difference is what the word generative marks.

Two kinds of task

Predictive

Input: an email, a photo, a transaction
Output: one label from a fixed set
Example: spam / not spam
Example: cat / dog / neither

Generative

Input: a request or description
Output: newly constructed content
Example: a reply to your question
Example: a picture matching your description

The same product often mixes both. A photo app may first classify an image (predictive) and then generate a cleaned-up version (generative). The label describes the task, not the whole product.