MEPX
Chapter 1 of 10All chapters

Chapter 1 of 10

What generative means

Producing, not classifying.

The difference

A classifier sorts inputs into categories. A generative model learns the shape of its training data well enough to produce new samples from it.

  • Novel output is the point, which is also why it can be confidently wrong.
  • The same underlying idea covers text, images, audio and code.

Not a database

It stores patterns in weights, not documents. Occasionally it reproduces training text closely, which is a memorisation failure rather than the normal mode.