Analysis
Lexicon
Lexicon · I · The model
LLM
Also: large language model
The engine: a statistical model trained on vast amounts of text, able to produce language. It is the thing every generative-AI product puts to work.
The analogy
The LLM is an engine. Powerful, staggeringly expensive to build, unusable as-is: nobody commutes sitting on an engine block. To do anything with it, you mount it in a vehicle: a consumer chat app, a development tool, an internal product. When an organization says "we have AI", the next question should always be: which engine, mounted in which vehicle, driven by whom.
A concrete example
GPT-5 (OpenAI), Claude (Anthropic), Gemini (Google) and Llama (Meta) are LLMs. The same Claude powers the consumer chat application, the Claude Code development tool, and hundreds of enterprise products calling it over an API. One engine, dozens of vehicles, and results that have nothing in common, because the vehicle matters as much as the engine.
What it is NOT
It is not a database: an LLM does not "store" facts to look up later, it produces the most plausible text, which is why it can state false things with total confidence. It is not a mind: it has no intent and no memory of its own. And it is not "AI" in general: it is the base brick, at the lowest level of the stack; everything else (access, orchestration, organization) is built on top.
Often confused with
The distinction that changes the conversation.
- Chatbot (ChatGPT, Claude…)
- The chatbot is a finished product; the LLM is its engine. ChatGPT is not a model: it is an application built around the engine of the moment (GPT-5 yesterday, GPT-5.5 today), the way a car is built around an engine. 'We use ChatGPT' and 'we use GPT-5.5' are two different statements: one names a product, the other an engine that can be mounted elsewhere.
- AI
- AI is the whole discipline: vision, robotics, recommender systems and much more. The LLM is one particular family: the one that produces language, and that changed everything after 2022. Not all AI is an LLM.