Forget the film version for a minute. No glowing red eye, no mind in a jar.
Ask an AI chat "Explain what a tattoo deposit is in plain English" and it writes something sensible. But it isn't thinking like a person, and it isn't looking up the one correct answer in some vault of all human knowledge — not the way most people picture it.
Next-word prediction, at massive scale
During training, the system read a colossal pile of text — books, websites, forums, manuals, the lot — and learned one thing extraordinarily well: after these words, these words tend to follow.
When you send a message, it builds the reply one chunk at a time, each time picking what best fits the pattern given everything written so far. Your phone already does a baby version of this — tap the middle autocomplete word over and over and you get a daft little sentence. Chat AI is that same trick, fed the entire internet instead of your last few texts.
That's the core of it. Pattern matching for language. Everything else — "reasoning", coding help, summaries, images, voice — is layers stacked on top of that one move. Genuinely powerful. Still not a soul in a box.
LLM — the jargon, translated
You'll see the letters LLM everywhere. It stands for Large Language Model: large (a huge training run), language (text in, text out), model (the pattern system doing the predicting). You don't need to remember the acronym. You need the mental picture: very advanced autocomplete that can hold a conversation.
For months I thought it was "Google running in its head." Dead wrong — and it cost me time, because I trusted it like a search engine. The day I started picturing autocomplete that sounds confident whether it's right or not, I wrote sharper prompts and stopped taking its facts on trust.
Continue — what it's actually good and bad at.