What is an LLM?

A large language model is a neural network trained on vast amounts of text to do one thing extremely well: predict the next token given everything before it. That single objective, repeated across enormous training corpora, forces the model to absorb grammar, facts, reasoning patterns, and styles, because guessing the next word correctly requires all of them.

Modern LLMs are built on the transformer architecture, whose attention mechanism lets the model weigh every earlier token when predicting the next one. Generation is then a loop: the model produces a probability distribution over the vocabulary, samples one token, appends it to the sequence, and predicts again. Settings like temperature shape how predictable that sampling is.

This design explains both the strengths and the failure modes. Text generation is flexible, so the same model summarizes, translates, and writes code. But a predictor of plausible text can produce fluent output that is simply wrong, the failure known as hallucination. Knowledge is also frozen at training time, which is why real systems pair the model with retrieval and tools.

Everything else in this series is a layer on top of that prediction loop.