What is an LLM?
A large language model (LLM) is a neural network trained on huge amounts of text to predict and generate language, used for drafting, summarizing, extraction, classification and reasoning over text.
In more detail
Models such as GPT, Claude, Gemini and Llama predict the next token in a sequence, and at scale that one skill covers a wide range of work. An LLM can read an invoice and return the fields as JSON, classify a support ticket or summarize a contract clause. It can also state invented facts with confidence, forgets everything between calls unless you pass it back in, and is billed by the token. In business systems the model is one component. Retrieval, validation, permissions and logging do the rest. Model choice is a cost and latency decision as much as a quality one: smaller models handle routine classification cheaply, and larger ones are kept for hard reads.
Example
A property manager runs every tenant email through an LLM that returns three things: the category (maintenance, billing or lease question), an urgency score and a draft reply. A coordinator reviews the draft before it goes out.