What is Generative AI
Gen AI · Foundations of Generative AI
Theory
Generative AI is a family of models that create new content (text, images, audio or code) instead of only labelling existing data. A traditional, discriminative model answers "is this email spam?"; a generative model answers "write a polite reply to this email." It learns the patterns in very large amounts of training data and then produces new output that plausibly follows those patterns, one small piece at a time.
For a data learner, the practical shift is that you can ask in plain language for a SQL draft, a summary of 500 customer reviews or a product description. The output is a likely answer, not a looked-up fact, so it still needs checking before you rely on it.
Why we use it
It automates work that used to need a human writer or analyst: drafting, summarising, translating, explaining and generating code. Tasks with no single correct label, like "summarise these complaints", become something software can do.
When to use it
Use it for open-ended writing, summarisation, pulling fields out of messy text, and assistants. When you need an exact figure from your warehouse or a yes/no prediction on structured data, plain SQL or a classic ML model is usually cheaper and more reliable.
How it works
- You give an instruction (a prompt)
- Model reads the prompt as tokens
- Model predicts a likely next piece of output
- Is the response complete? Yes: Stop and return the output. No: Append the piece and predict the next one.
- You review and use the generated content
Example
Prompt: "Write a 2-line product tagline for a reusable coffee cup."
Output:
"Your coffee. Your planet. One cup at a time."
(A discriminative model, given the same cup, could only tell you its brand or color — it can’t write new copy about it.)Key points
- Learns patterns from training data and samples new content from them
- Discriminative AI predicts a category or number; generative AI creates new content
- Output is a plausible answer, not a verified fact: check important claims
- The same idea covers text, images, audio and code
- Powers chat assistants, image generators and coding assistants