What is Generative AI?
You'll be able to
- Define generative AI in contrast to discriminative models
- Understand the concept of a probability distribution
- List what modern generative systems can create
Generative AI is a class of models that learns to create new content — text, images, audio, code — by modelling the distribution of its training data. Instead of classifying an image as 'cat', it can produce an entirely new image of a cat.
Under the hood, a generative model learns what 'likely' data looks like, then samples from that learned distribution to produce novel, plausible outputs.
What it can create
- Text: articles, summaries, code, dialogue.
- Images: photoreal pictures from a description.
- Audio: speech, music, sound effects.
- Structured data: SQL, JSON, plans.
Generative models sample from probability, so outputs vary run to run — a feature for creativity, a caution for correctness.
Challenge
Generate, don't classify
List three tasks where you would want a generative model and three where you would want a fixed, deterministic classifier.
Knowledge Check
Generative AI
The defining ability of generative AI is to:
A generative model typically gives the exact same output every time you ask it the same question.
Answer all questions to submit.