LLM stands for Large Language Model, and the name is more honest than most of the marketing around it. It is a program that learned language by reading a huge slice of the internet. Its whole job is predicting the next word, and then the next one, well enough that the result looks like thinking. There is no encyclopedia inside checking facts, and no rulebook of logic. There is one extremely well practiced instinct: given everything so far, what text comes next?
That single fact explains almost everything the model does, good and bad. It is why the output reads so naturally: the model has seen more writing than any human ever will, and it continues yours in kind. It is also why it can state something false without blinking. Plausible is what it optimizes for. True is not a concept it holds. The polish of the prose tells you nothing about the accuracy of the content.
For an operator, the practical consequence is that wording becomes a control surface. The same model produces a throwaway answer or a genuinely useful one depending on what you feed it: the instructions, the examples, the context, the tone. Teams that treat the model like a search box get search-box results. Teams that treat it like a very fast, very literal junior colleague, one who needs clear direction and whose work gets reviewed, get real leverage out of it.
A concrete example. A small firm answers the same customer questions over and over. An LLM drafts each reply in the company's voice, working from the message and the instructions it was given, and a person reviews before anything sends. The model never decides policy and never touches the order system. It predicts what a good reply looks like, which is exactly the job it was built for, and the hours it hands back are the point.
“The model is just predicting text, so how we word things changes what we get back.”
// this page is the full text. the packaged PDF, all fifteen terms, comes with the free library.