What AI actually is
ChatGPT, Claude, Gemini — what's underneath the hood.
By the end of this lesson
After this lesson, you'll be able to explain in one sentence what an AI like ChatGPT is, in language your mum would understand.
Most of us have used ChatGPT, Claude, Gemini, or Copilot at least once by now. We've watched it answer a question and felt either impressed or unsettled. Sometimes both.
What's harder is explaining what it actually is — to a friend, to your mum, to your eight-year-old. The honest version isn't complicated. It just hasn't been said clearly enough.
An autocomplete that read a huge amount
Think about the autocomplete on your phone. You type "I'll see you at the" and it suggests "station", "office", or "park" — based on patterns it's seen before in your typing.
ChatGPT is the same idea, just enormously bigger. Instead of learning from your texts, it was shown a huge mix of text from the internet — books, websites, Wikipedia, forums, news. It learned the patterns: which words tend to follow which, given everything that came before.
When you ask it a question, it generates an answer one word at a time. Each next word is chosen because it's the most likely continuation of what came before. Scaled up enormously, that simple trick produces answers that look like reasoning.
Like the autocomplete on your phone — but trained on a huge mix of books, webpages, and Wikipedia-style text.
It doesn't know. It predicts.
Here's the part that surprises people. The model doesn't store facts the way a database does. It stores patterns of words that tend to appear together.
So when you ask "what's the population of New Zealand?" it isn't looking up a number. It's predicting what number tends to follow that question in the patterns it learned. Sometimes the prediction is right. Sometimes it's confidently wrong.
This is why a tool that sounds like the smartest person in the room can also tell you the capital of Australia is Sydney — and not blink. It doesn't know. It predicts.
If a friend confidently told you something but couldn't show you where they got it, you'd ask "where did you read that?". Same friend, same question, applies to AI.
When was the AI's last day at the library?
Every AI has a training cutoff — a date when it stopped reading new information. Anything after that date, the model doesn't know. New films, new prices, new world events, the people you've met since.
If you ask "what happened at last week's Cabinet meeting" to a model trained two years ago, it'll either admit it doesn't know — or confidently make something up.
This is fixable: some tools (ChatGPT with browsing, Claude with web search, Copilot at work) can search the live web and bring back current sources. When they do, they should show you those sources. If they don't show sources, treat the answer as a guess.
Imagine someone who left for a year-long ocean voyage in 2023, just got home, and is confident about everything — including current events. That's an AI without web search.
Try this for 2 minutes
Open ChatGPT, Copilot, Gemini, Claude, or any AI chat tool you can access for free. Ask it two questions:
- "What's the capital of Australia?"
- "What's my dog's name?"
The first one it'll answer confidently. The second it won't know — it might guess, or admit it can't help. Notice: same model, but it can't tell you what it can't tell you.
That's the limit you've now seen with your own eyes.
Keep this
- Rule — AI predicts likely-sounding answers. It doesn't search live facts unless the tool shows sources.
- Phrase — "Where did you get that?" — say it to AI as often as you'd say it to a friend.
- Don't — Don't trust an AI fact you can't trace back to a source.
Pop quiz, no marks
-
If your nephew says "ChatGPT KNOWS the answer to that", what's the friendly correction in 10 words?
Show answer
It doesn't know — it predicts a likely-sounding answer.
-
What's the rule when an AI gives you a date or a number?
Show answer
Look for a source. If there isn't one, treat the number as a guess.
Want to go deeper?
Each of these topics on Plain AI explores one idea from this lesson in more detail:
Sources
- Stephen Wolfram — What is ChatGPT doing and why does it work? [third-party]
- Wikipedia — Large language model [third-party]
- OpenAI — Introducing ChatGPT [vendor]