What AI actually does — Foundations

Every answer from ChatGPT, Claude, Gemini or Grok is built one small piece at a time. The model looks at everything written so far and predicts what should come next — then does it again, and again, thousands of times a second.

That's it. No database lookup, no little person inside. Just an astonishingly good prediction machine called a large language model (LLM).

To get good at predicting, the model was trained on a huge slice of human writing — websites, books, code, conversations. To predict the next word of a physics explanation, it had to pick up physics. To finish Python code, it had to learn Python.

Knowledge, reasoning and style all emerged as side effects of getting really, really good at guessing what comes next.

Because it predicts rather than retrieves, an AI is great at: drafting, rewriting, summarizing, explaining, brainstorming, translating, coding, and reasoning through problems.

And it's risky at: exact quotes, niche facts, recent news (unless it can search), precise citations and numbers you haven't given it. Keep that split in your head and you'll already use AI better than most people.