What AI actually is
In one line: the AI you talk to is a program that is extremely good at guessing what comes next, and everything it can and cannot do follows from that one fact.
The short version
Someone showed a program an enormous amount of writing — books, code, articles, arguments, recipes. Not to memorise it, but to get good at one game: given some text, guess the next bit.
Play that game well enough and something surprising happens. To guess the next word of an explanation, you have to have picked up how explanations work. To guess the next line of a program, you have to have picked up how programs work. Nobody taught it the rules on purpose. They fell out of getting very good at guessing.
That is a large language model, and it is what is behind ChatGPT, Claude, Gemini and the rest.
Why this explains almost everything
Once you hold on to "it is guessing what comes next", the odd behaviour stops being odd.
It is confident even when it is wrong. It is not looking anything up. It is producing the most plausible continuation, and a plausible-sounding wrong answer looks exactly like a plausible-sounding right one from the inside. This is usually called hallucination. It is not lying — there is no intention in there — it is guessing, and sometimes the guess is wrong.
It is much better when you give it more to go on. A guess made from three words is worse than a guess made from three paragraphs. This is the single biggest thing you can do to get better answers, and it is what AIOS is doing for you when it sends your notes along with your question.
It does not remember you. Each conversation starts cold. Anything it "knows" about you was in the text you just sent it. When a tool seems to remember, something is quietly re-sending your history each time.
It has a cut-off. It was trained on text up to some date, and it does not know about anything after that unless the tool gives it a way to search.
What it is not
It is not a database, a search engine, or a person. It has no goals, no opinions it holds, and nothing at stake. When it says "I think", that is the shape of the writing it learned, not a report from inside.
That is not a criticism. A very good guesser of what comes next turns out to be extraordinarily useful. But you will get far more out of it if you are not expecting something else.
Try this
Open the prompt box in AIOS and ask it something you already know the answer to — something from your own work, where you would spot a mistake instantly. Read the answer as a draft by a fast, well-read assistant who has never met you. That is the right posture, and it is the one that gets the most out of these tools.
If you want to go further
- Anthropic — Introduction to Claude
- OpenAI — What are tokens
- Google — Introduction to large language models
---
Next: Who makes the models · All lessons