What is a large language model?

An LLM learns patterns in language from enormous amounts of text, then uses them to work out what to write next, a word at a time. That simple idea, at huge scale, lets it summarise, draft, translate, answer questions, read documents, write code and follow instructions.

An example. A distributor receives orders as emails, PDFs and spreadsheets in every layout imaginable. An LLM can read each one, pull out the customer, products, quantities and delivery date, and hand them to the order system to check, a job that used to mean rekeying.

What it's good at

Anything that involves reading or writing language: drafting replies and product content, summarising long documents and accounts, classifying incoming emails and cases, extracting details from documents, and powering assistants that answer questions in plain English.

What to know before you rely on one

• It doesn't know your business. It knows what it learned in training, up to a date. To answer from your products, prices and policies, it needs to be given them, usually through RAG.
• It can be confidently wrong. It writes what sounds right, which is usually, but not always, what is right. Answers that matter need grounding in your sources and a way to check.
• It can be steered by what it reads, which is why AI that acts on your systems needs protecting from prompt injection.
• Your data needs the right terms. Business versions of the main models can be set up so your data isn't used to train them. Check the licence before staff paste in customer information.

Choosing between the main models for your business: Claude vs Copilot vs ChatGPT Enterprise for UK businesses, from Enceladus AI, the PureNet Group's AI strategy company.