Foundations
Large language model (LLM)
A model trained to predict the next piece of text, which turns out to be enough to summarise, draft, classify and reason about language.
An LLM is trained on a very large amount of text with one deceptively simple objective: predict what comes next. Scale that up far enough and the model becomes able to answer questions, write code, extract data from documents and hold a conversation.
Two consequences follow, and both matter to a business:
- It is generating, not looking up. An LLM has no database of facts to consult. It produces plausible text. Usually that text is correct; sometimes it is confidently wrong. See hallucination.
- It has no memory between conversations. Everything it "knows" about your business has to be supplied each time, inside the context window.
Almost every design decision in a production AI system follows from those two facts.