Generative AI creates new content — text, code, images, audio, or video — from a prompt. It does not only label or score something that already exists.
Generative versus discriminative
| Approach | Question it answers | Example |
|---|---|---|
| Discriminative | Which class does this belong to? | Is this email spam? |
| Predictive | What number comes next? | What is the expected loss on this loan? |
| Generative | What new content should I produce? | Draft a reply to this customer. |
How a text model works
A large language model is trained to predict the next piece of text, called a token, given everything that came before it. After training on a very large corpus, that simple objective produces fluent writing, translation, summarization, and code. You steer the result with instructions (a prompt), with your own documents (retrieval), or by further training (fine-tuning).
Where it helps
- Turn a long report into a short brief.
- Draft code, SQL, or a first version of an email.
- Answer questions from a policy manual when paired with retrieval.
- Explain a model result in plain language for a non-technical reader.
A practical example
An analyst has a 40-page credit memo. A generative model can produce a one-page summary, list the open risks, and draft the risk paragraph in the team’s usual tone. A person still checks the numbers. The model drafts; it does not own the decision.
What to remember
- Generative AI creates. Classification and regression score or predict.
- Fluency is not the same as being correct. Check facts that matter.
- Prompts, retrieval (RAG), and fine-tuning are the three main ways to adapt a general model to your work.
