What the thing actually is

A large language model is one very large machine for guessing what text comes next.

Part of the What is this thing? track on lAItest.

You typed a question and something typed back. Nothing in there looked anything up.

So where did the answer come from?

It is a guessing machine for text.

A large language model — LLM for short — is one enormous mathematical function. Text goes in. The function produces a score for every possible next piece of text. High score means likely. That is the whole machine. Everything it appears to do — answer, translate, write code, argue — is that one operation, run over and over.

A common misconception

Commonly believed: Inside there is a program that understands the question, searches for the answer, and writes a reply. Something like a very fast librarian.

Actually: There is no search step and no shelf of answers. The model was shaped by a mountain of text until it became extremely good at continuing text. When it continues your question, the continuation is often a correct answer. Often is not always, and that gap is most of what this track is about.

Why "large"?

Large refers to the numbers inside. A model is a pile of numbers, called parameters, that decide how input turns into output. The big ones hold hundreds of billions of them, and some hold more. Nobody wrote a single one by hand. They were set by a process called training, which gets its own lesson later in this track.

In one sentence

A large language model is not a librarian and not a search engine. It is a machine that got extremely good at one narrow trick: given some text, produce plausible next text.