Back to Lessons
beginner10 min10 min read

How AI Actually Works: Autocomplete on Steroids

No jargon, no math, no PhD required. Learn what a large language model really does under the hood - and why it's both impressive and slightly dumb.

What you will learn

  • Explain how large language models work using the autocomplete metaphor
  • Understand what training means without any math
  • Explain why hallucinations happen and how to guard against them
  • List what AI is genuinely good at versus what needs verification

The Best Mental Model You Will Ever Get

Forget everything movies taught you about AI. No robot is plotting anything in there. The honest description of today's AI is: an extremely advanced autocomplete.

When you type into ChatGPT or Claude, the model's only real job is to predict the next word - over and over again, millions of times per second. Type "The cat sat on the..." and it guesses "mat". That's the whole trick. The magic is that it does this at a scale and speed that makes it look like it's thinking.

How the Model Was "Trained"

Imagine reading every book, article, forum post, and recipe the internet has ever produced - then summarizing the patterns into a giant set of numbers called weights. That's training.

  • The model read billions of pages of text
  • It learned which words tend to follow which other words
  • It stored that knowledge as numbers, not as facts in a database
  • The result is a system that can generate text that sounds like a human wrote it

Here's the catch: the model doesn't know anything the way you do. It has never seen a cat. It just knows that "furry" and "whiskers" often appear near "cat" in the text it read.

Why It Hallucinates

Because the model is predicting words, not looking things up, it sometimes invents confident-sounding nonsense. That's called a hallucination - and it's not a bug, it's the core design.

  • It will happily invent a book that never existed
  • It will cite a court case that doesn't exist
  • It will be 100% sure about all of it

This is the single most important thing to know about AI: it is a confidence machine, not a truth machine. Treat everything it says as a suggestion from a very articulate intern who needs fact-checking.

What It's Actually Good At

  • Drafting: emails, posts, plans, first versions of anything
  • Explaining: ask it to explain something like you're ten
  • Brainstorming: it never gets tired of bad ideas, and occasionally has good ones
  • Summarizing: feed it long text, get the short version
  • Formatting: turn scattered notes into a tidy table or list

Try This Today

Open any free AI chatbot and ask it: "Explain how you work in one paragraph, using a sandwich metaphor."

Then ask it: "Now give me three facts about yourself, and mark which one is most likely to be made up." You will learn more about AI from that exchange than from a week of reading.

The One-Line Summary

AI is a statistical parrot with an excellent vocabulary. Useful beyond belief, but it needs a human fact-checker on the payroll.

ai-basicsllmhow-ai-workshallucinationsgetting-started
🤖

Almost Done!

Made it to the end — nice work. Record your achievements to update your smart-assistant profile.

Scroll progress: 0% • Finish reading down to complete.