LLM (large language model)
LLM (large language model): A large language model (LLM) is an AI model trained on enormous amounts of text to predict the next word, which lets it understand questions and write, summarize, translate, classify and reason about text.
An LLM, short for large language model, is the type of artificial intelligence behind ChatGPT, Claude, Gemini and Copilot. It trains on massive amounts of text to learn which word comes after another. At enough scale, that ability becomes very useful: understanding what you ask, writing, summarizing, translating, extracting data from a document or following multi-step instructions.
How does a language model work?
An LLM works with tokens, fragments of words. When you type, it converts the text into tokens, processes them and generates the answer token by token, picking the most probable one at each step given the context.
Three concepts help explain its abilities and limits:
- Training: the model learns from public text up to a cutoff date. It knows nothing later and nothing private about your company unless you provide it.
- Context window: the amount of text it can consider at once (your question, the instructions, the documents you pass in). Current models handle long documents, but not a whole company.
- Hallucinations: because it generates probable text, it can produce facts that sound right and are false. You reduce this by giving it the right information with techniques like RAG and asking it to say when it does not know.
Which language models exist?
The ones most used in business come from a few providers:
| Family | Company | Type |
|---|---|---|
| GPT | OpenAI | Commercial, via API and ChatGPT |
| Claude | Anthropic | Commercial, via API and the Claude app |
| Gemini | Commercial, via API and Google apps | |
| Llama | Meta | Open weights, can run on your own server |
| Mistral | Mistral AI (France) | Commercial and open-weight models |
Open-weight models appeal when data cannot leave your infrastructure, in exchange for running the server yourself.
What does a business use an LLM for?
On its own, an LLM is a good writer. Its value in a business appears when it connects to processes and tools:
- Customer service: answer email, WhatsApp or chat with the company’s information.
- Documents: read invoices, contracts or delivery notes and extract the data into a system.
- Classification: sort incoming email, tag tickets or detect how urgent a message is.
- Writing: quotes, product sheets, replies to reviews or meeting summaries.
- Agents: the LLM is the “brain” of an AI agent, which also uses tools to act.
Example: an ecommerce store with hundreds of emails
Picture an online pet supplies store that receives about 150 emails a day: shipping questions, returns, product questions and the odd complaint. Two people read every one to decide who replies.
The store connects an LLM to its inbox. For each email, the model identifies the type of request, extracts the order number if there is one and detects whether the customer is angry. Shipping questions get answered automatically with the real order status; returns are prepared as drafts for review; complaints go straight to a person with a summary. The team stops sorting and spends its time on the cases that need it.
How do you choose and get started?
You do not need to pick one model forever. Design your automations so you can switch providers when a better or cheaper one appears, and use different models for different tasks. To find where to apply an LLM in your company and at what cost, start with an AI consulting engagement.