what sign up
nixm nixm

← learn

What is generative AI? The plain-English version

October 14, 2026

Generative AI is software that produces new content, text, images, video, sound or code, from a prompt you write. It works by learning patterns from enormous amounts of existing material and then predicting, piece by piece, what a plausible answer looks like. Older AI classified and sorted things you gave it; generative AI makes things you asked for. That is the whole difference, and it is why it feels new.

The one-sentence version

Generative AI is a program that, given a description, produces something that did not exist a second ago: a paragraph, a picture, a short video, a voice, a spreadsheet formula, a working web page. "Generative" just means it generates. "AI" is the fact that nobody wrote the rules for how; it learned them from examples.

How it actually works, without the maths

Take a model that writes text. It was trained by reading a very large amount of writing and, over and over, guessing the next word, then checking the guess against the real next word, then adjusting. Do that enough times and the model ends up with a dense, compressed sense of how language goes: which words follow which, how an argument is shaped, how a recipe is laid out, how a legal clause sounds. When you type a prompt, it continues from there, one small piece at a time, each piece chosen because it is likely given everything before it.

Image, video and audio models do the same thing with pixels, frames and sound instead of words. They learn from a huge pile of captioned pictures, or clips, what "a fox in snow at dusk" tends to look like, and then produce a new one that fits the description. It is not retrieving a stored picture. It is making one that is statistically like the ones it learned from.

The key word is likely. A generative model does not know things; it produces plausible things. That is why it can write a confident paragraph about a court case that never happened. Plausible and true overlap most of the time, which is what makes the tool useful, and not all the time, which is what makes it dangerous to trust blindly.

Generative AI vs the AI we already had

AI has been in your life for years: the spam filter, the map that reroutes you, the bank's fraud flag, the photo app that finds your dog. All of that is discriminative AI: it looks at something and decides what it is or what to do with it. It sorts.

Generative AI inverts that. Instead of "here is an email, is it spam?", it is "write me an email". Instead of "which of these photos has a dog?", it is "make me a photo of a dog". Same family of techniques, pointed the other way. The reason 2022 felt like a line in the sand is that this inversion happened to land on language, and language is the interface to almost everything people do at a desk.

What "gen AI" can make today

  • Text: drafts, summaries, translations, explanations, code. The most mature kind.
  • Images: from a sentence, in nearly any style, in seconds.
  • Video: short clips, often with sound generated at the same time. A few seconds per clip is still the norm; a minute is unusual.
  • Audio: voices, music, sound effects.
  • Structured things: a chart from pasted numbers, a slideshow from an outline, a mind map from a brain dump, a small working app from a description.

Most people meet it through a chat window, because a chat window can ask for any of those. In nixm, for example, you type what you want in one place and get the paragraph, the chart, the image or the clip drawn back into the same conversation, which is a fair picture of where this is going: one place to ask, many kinds of answer.

What it is good at

First drafts. Summaries of things too long to read. Explaining a topic at whatever level you ask for. Translating across languages and registers. Producing many variations fast so you can pick. Writing the boring half of code. Turning a description into an image when you cannot draw. These are the jobs where "plausible" is enough and you are there to judge.

What it is bad at

Knowing when it is wrong. Counting. Keeping a long document consistent. Anything that needs a fact it was never shown. Staying private, by default, since most consumer tools keep what you type and many train on it. The limitations article goes through each one, and the pros and cons page weighs them against the upside.

Three things to know before you use it

  1. Check anything that matters. Treat a factual claim from a generative model as a lead, not a source. Ask for the source and open it.
  2. Know where your words go. Most tools keep your conversations. Some train on them. Search the privacy policy for "train" before pasting anything personal, or use a tool that does not keep a transcript at all.
  3. Use it to start, not to finish. The draft is the model's; the judgement is yours. People who use it that way get the speed without losing the skill.

frequently asked

What does generative AI mean?

Software that generates new content, text, images, video, audio or code, from a prompt, by predicting what a plausible answer looks like based on patterns it learned from a large amount of existing material.

What is the difference between AI and generative AI?

Traditional AI classifies and decides: is this email spam, which photo has a dog. Generative AI produces: write the email, make the photo. Same underlying techniques, pointed at creating instead of sorting.

What are examples of generative AI?

Chat assistants that write and summarise, image generators, short-video generators with sound, voice and music generators, and tools that turn a description into a chart, a slideshow or a small app.

Is ChatGPT generative AI?

Yes. ChatGPT, Claude, Gemini, Copilot and similar chat assistants are generative AI for text, and several of them also generate images.

Does generative AI know facts?

No. It produces likely text, which is usually accurate and sometimes confidently wrong. Check anything that matters against a real source.

Is generative AI safe to use?

For most everyday tasks, yes, with two habits: verify factual claims, and know what the tool keeps. Many consumer tools store and train on conversations unless you turn that off.

try it now — no account, no transcript kept. even the signal dissolves in seven days.

seed a signal →

Enter your email and receive a link to sign in. No passwords, just a secure signal to connect.

why connect

  • 20 free credits to start
  • your signals, remembered
  • buy more credits anytime
  • image & video generation features
or
+