“Generative AI” gets used as a catch-all term for almost anything AI-related right now, which makes it hard to tell what’s actually new about it. Here’s the practical distinction.
Traditional AI Predicts or Classifies
Most AI systems built before the last few years were designed to answer one narrow question well: is this transaction fraudulent, what’s in this image, will this customer churn. They take an input and return a label, a score, or a prediction. That’s genuinely useful, but the output is always drawn from a fixed, predefined set of possibilities.
Generative AI Creates New Output
A generative model, by contrast, produces new content — text, code, images — that didn’t exist as an option beforehand. Ask it to draft an email, summarize a document, or write a function, and it constructs a fresh response token by token, rather than picking from a list. That’s the shift that made tools like chat-based assistants possible: the system isn’t matching your question to a canned answer, it’s generating one.
Why This Distinction Matters Practically
If you’re evaluating where AI fits in your work, the question isn’t “is this AI or not” — almost everything is now. The useful question is whether you need a classifier (a fixed decision from fixed categories) or a generator (open-ended content production). They’re built differently, evaluated differently, and fail differently. A classifier fails by picking the wrong category. A generative model can fail more subtly — producing fluent, confident output that’s simply wrong, which is why grounding it in real data (retrieval) and reviewing its output matters more than it does for a classifier.
Understanding that difference is the first real building block for working with generative AI seriously, rather than just using it as a novelty. It’s also exactly where our Generative AI course starts.