Generative AI is a class of artificial intelligence that creates new content — text, images, audio, video, or code — by learning patterns from large…
Define generative AI as models that create new content from learned patterns.
Generative AI is artificial intelligence that creates new content — text, images, audio, video, or code. Traditional (discriminative) AI answers 'which category is this?'; generative AI answers 'produce something new like this.'
It learns statistical patterns from huge datasets, then samples from those patterns to generate original outputs that resemble the training data without copying it.
Most generative AI runs on foundation models — large neural networks pre-trained on broad data (all of the web's text, billions of images). Examples: GPT and Claude for text, Stable Diffusion and Midjourney for images, Whisper for speech.
Explain prompt-in, content-out and why the same prompt can give different results.
You give the model a prompt — 'write a haiku about the sea' or 'a photo of a red bicycle' — and it generates the content one piece at a time, each piece chosen from a probability distribution the model learned.
Because it samples from probabilities, the same prompt can produce different outputs each time. That creativity is a feature, but it also means results aren't perfectly repeatable.
A text model generates the next likely word given everything so far. It has no database of facts to look up, which is why it can sound confident yet be wrong — a failure called hallucination.
Cover the modalities of generative AI with concrete examples.
Text: chat assistants, summarizers, translators (GPT, Claude). Image: text-to-image art and editing (Stable Diffusion, Midjourney). Audio: speech synthesis and music. Video: text-to-video (Sora, Veo). Code: programming assistants (Copilot, Claude Code).
Multimodal models handle several at once — read an image and answer questions about it, or turn a text script into a narrated video.
Give an honest capability boundary and one caution.
Generative AI excels at first drafts, brainstorming, summarizing, translating, and boilerplate code. It struggles with exact facts, current events beyond its training cutoff, arithmetic, and anything requiring guaranteed correctness.
Treat it as a fast, fluent assistant whose output you verify — not an oracle.
Generative AI creates new content — text, images, audio, video, code — by sampling from patterns learned by a foundation model. You steer it with a prompt; it predicts likely output, which makes it creative but also prone to confident errors, so you verify what it produces.
Pick one everyday task you'd want help with. Which modality of generative AI fits it, which tool would you try, and what would you double-check in the output before trusting it?
What is generative AI?
Generative AI produces original content; discriminative AI only classifies. That creation is the defining trait.
Why can the same prompt give different outputs?
Sampling from probabilities gives varied, creative outputs — repeatability isn't guaranteed unless you fix the randomness.
What is generative AI weakest at?
It's strong at fluent drafts and ideas, weak at exact truth — which is why you verify its factual output.