Large Language Model (LLM): Definition
What a large language model is
A large language model, or LLM, is an artificial intelligence model trained on huge amounts of text to predict the next word. More useful abilities grow out of that simple one: reading a document and summarizing it, answering questions, extracting data, translating or writing a draft.
It is called large because of the volume of text it is trained on and the number of parameters it has, the internal values it adjusts during training.
How a large language model works
An LLM does not look up answers in a database. During training it learns patterns in language: which words tend to go together, how an argument is structured, how a contract is written. When it receives an instruction, it generates the answer word by word based on those patterns.
This has two practical consequences:
- It writes fluently, even on technical subjects.
- It can be confidently wrong. If it does not have the information, it can make it up. That is why, in a business, it should work from specific documents and each answer should show its source.
LLMs for business
In business, LLMs are used mostly for work with documents and text:
- Reading long documents and summarizing what matters.
- Extracting specific data from contracts, invoices or tender specifications.
- Answering questions about internal documentation.
- Writing drafts that a person then reviews.
Results are better when the model receives the relevant documents with each query, using techniques such as RAG, instead of answering from memory. And when a person reviews what it produces before it is used.
A simple example
A company receives a tender with several documents and dozens of pages. An LLM connected to those documents can prepare a summary sheet with the budget, deadlines, required qualifications and penalties, pointing to the page of each item. The bid specialist reviews the sheet and decides whether to bid, instead of reading everything from scratch.