Large language model (LLM)

A large language model is a model trained to predict the next token in a sequence. Scaled up far enough, and then tuned to follow instructions, that one skill turns into something broadly useful: it can answer questions, rewrite text, pull fields out of an email, classify a ticket, or decide which function to call.

As a builder you almost never train one. You call it through an API (or run an open-weight model yourself), send it instructions and context, and get text or structured data back. Everything the model knows about your users and your data arrives in that request, because the model has no memory of its own between calls.

The mental shift that matters is that an LLM is probabilistic, not deterministic. The same input can produce different output, it can be fluent and wrong at the same time, and its quality varies by task. That is why AI features need schemas, validation, grounding in your own data, and evals rather than a single manual test.

Choosing an LLM is a per-feature decision. A high-volume ticket classifier and a hard multi-document analysis rarely belong on the same model, and the price and latency gap between tiers is large enough to shape your margins.