LLM
A large language model — a neural network trained on huge amounts of text that predicts the next chunk of text given what came before.
An LLM is a specific kind of neural network trained to predict the next token (roughly the next word) given a sequence of prior tokens. The training material is typically hundreds of billions of words scraped from the web and licensed sources; the model has billions to trillions of numerical parameters that get adjusted during training.
The reason LLMs can do things beyond next-word prediction — writing essays, translating, coding, reasoning through problems — is that "predict the next token well across huge amounts of text" turns out to require developing something functionally like general knowledge and reasoning as an intermediate step. Whether that "something" is understanding in a meaningful sense is a live philosophical debate; whether it's useful in a practical sense is settled.
"LLM" as a term is slightly imprecise. Modern frontier systems combine language modeling with other components (image and audio processing, tool use, retrieval), so calling them all just "LLMs" flattens meaningful architectural differences. The industry hasn't settled on a better catch-all yet.