05What an LLM is, plainly

The most useful thing you can do early on is stop thinking of AI as either magic or a person. It is neither.

Modern AI is a language model. It read an enormous slice of everything people have written and learned exactly one thing: guessing what comes next. You write the beginning, it continues. Everything else is built on top of that.

Sounds dull, but all of its power grows from there. It has read so much code, documentation and discussion that "just continue the text" means writing a working function, finding a bug, or explaining why the server isn't responding.

Its weaknesses grow from the same root. It doesn't "know" in the human sense and doesn't check itself. Point it towards nonsense and it will cheerfully continue towards nonsense.

Understanding this cures two fears at once. First: "what if it's judging me" - it isn't, ask anything. Second: "it's smart so it must be right" - also no, verify.