Putnam Valley Free Library

What they mean by large language model

You will hear the term large language model, or LLM, and it sounds like something requiring an engineering degree. It does not. Here is the entire idea, in kitchen English. Take an enormous amount of everything humans have written, books, articles, recipes, arguments. Let the computer read it, not for meaning exactly, but for patterns. Which words tend to follow which words. Then hand it a sentence and ask it to continue in the most sensible way, one piece at a time. That is the whole machine. It is a finish the sentence device with a library inside it, trained on more reading than any person could do in a thousand lifetimes. This explains everything you need to know about how it behaves. It writes fluently because it has seen a billion fluent sentences. It sometimes invents facts because a plausible sounding sentence and a true sentence look identical to a pattern machine. It is polite because the reading included a great deal of polite writing. When your grandson says the model has 400 billion parameters, he is describing the size of the pattern net, the same way you might describe a brain by its folds. Bigger net, more patterns caught. Useful to know, boring to memorize. The working knowledge is this. You are talking to something that has read everything and understood nothing, and if you keep that in mind, you will use it well and be surprised by it less.

Want the hands-on version? Come try it at a free class →