LLM. The engine inside AI assistants, explained without jargon.

An LLM (large language model) is a program trained on huge amounts of text to predict the next piece of text, and therefore able to write, summarise, translate and answer in natural language. It is the engine inside assistants such as ChatGPT, Claude and Gemini.

This entry is part of the glossary of AI and automation, where every term has a short definition. Here the definition widens: how an LLM works, the four words needed to understand it, what it can do in a small business, where it goes wrong and what European law says about it.

What is an LLM

An LLM is an artificial intelligence model trained on text, with billions of parameters, that produces new text from what it receives. «Large» refers to size: the amount of text used to train it and the number of parameters, the internal values adjusted during training.

The architecture almost every LLM is based on is the transformer, proposed in June 2017 in a paper by Ashish Vaswani and colleagues, titled «Attention Is All You Need». The key mechanism is attention: the model decides, word by word, which other parts of the text to look at.

How it works, in four words

Four words are enough to understand what happens inside an LLM, and each has a practical consequence for whoever uses it in a business: how much it costs, what it knows, what it keeps in mind during a task, how repeatable the answer is.

Word	What it means	Why it matters to you
Tokenthe unit of text	The piece of text the model reads and writes: a word, part of a word, a symbol.	Usage is paid per token, and the length of a document is measured in tokens.
Trainingwhen it learns	The phase in which the model learns from text by adjusting its parameters. It happens beforehand, once, and it is expensive.	The model does not know what happened afterwards, nor your documents.
Context windowwhat it keeps in mind	How much text the model considers at once: the question, the attached documents, the conversation.	What does not fit in the window does not exist for the model.
Temperaturehow much it varies	The parameter that sets how predictable or varied the answer is.	For business work it is kept low: less imagination, more repeatable answers.

What it can do in an SME

According to Istat, among Italian enterprises using artificial intelligence in 2025, 70.8% use it to extract information from text documents and 59.1% use generative AI, written, spoken or for images. These are the two jobs where an LLM delivers most: reading and writing.

In practice an LLM reads an email in any language and understands what it asks, extracts the data from an order or an invoice, drafts a reply, a product sheet or a quote, summarises a long document, translates while keeping the tone. The decision whether the draft is right stays with a person.

Where it goes wrong

An LLM predicts the likely text, and likely text can be false. When it does not know the answer it tends to produce a plausible one anyway: that is a hallucination. In an article of 5 September 2025 OpenAI attributes it to training and evaluation procedures that reward guessing over acknowledging uncertainty.

The second limit is the date. The model knows the world up to its training, and does not know your documents. Both limits are contained by giving the model the right sources at the moment of the question, with the technique called RAG, and by putting a person where errors are costly: that is human in the loop.

LLM, chatbot and agent

The LLM is the engine; chatbot and agent are two ways of using it. A chatbot puts the model in a conversation and produces answers. An agent gives it tools to act, such as reading an inbox or writing into management software. The difference between the two is explained on the page AI agent or chatbot.

What the AI Act says

The AI Act, in Article 3, point 63, defines the «general-purpose AI model»: a model trained «with a large amount of data using self-supervision at scale», displaying significant generality and capable of performing a wide range of distinct tasks. Large LLMs fall under it, and the main obligations fall on those who develop them.

For a company that uses them, two rules already in force matter most. Staff AI literacy, in Article 4, from 2 February 2025, explained on the page about the training obligation. And transparency towards whoever talks to a system, in Article 50, from 2 August 2026.

How Itria uses it

Itria uses LLMs inside systems with rules written beforehand, never on their own. In Inbox AI the model reads requests and prepares a draft: on a test bench on 4 September 2026, 100 requests across 4 channels and 5 languages, the draft was ready in a median of 7.290 seconds, with the right language in 97 cases out of 100.

The same models prepare multilingual drafts and extract data in document entry. In every case whatever commits the company goes through a person before sending, and the list of models used is available on request, as written on the AI transparency page.

Related terms

AI hallucinations

The false statement made with confidence by a model. The best-known limit of LLMs, and the easiest to contain.

RAG

The technique that gives the model the right documents at the moment of the question, so that it answers from them.

Prompt

The written instruction given to the model. The quality of the answer depends a lot on how it is written.

Generative artificial intelligence

The family of systems that produce new text, images or sounds. LLMs are the part of it that writes.

Questions and answers