AI glossary Models & Architecture
What is a large language model (LLM)?
LLM (Large Language Model)
← All glossary termsA neural network trained on very large amounts of text to predict the next word, which lets it write, summarise, translate and answer questions.
Explained
A large language model (LLM) is a neural network, usually based on the transformer architecture, trained on vast text corpora to predict the next token in a sequence. Doing that at scale gives it a broad ability to generate coherent text, follow instructions, answer questions, summarise, translate and write code. Well-known families include GPT, Claude, Gemini, Llama and Mistral. On its own an LLM only knows what was in its training data, so in business it is usually combined with retrieval of company documents, tools and guardrails.
How an LLM works
An LLM splits text into tokens (pieces of words) and learns, during pre-training, to predict which token comes next. The transformer architecture lets it weigh every part of the input against every other part, which is what makes long, coherent answers possible.
After pre-training, the model is usually tuned to follow instructions and to behave safely, often with human feedback. When you use it, it generates the answer one token at a time, based on the prompt and on whatever context it is given.
What businesses use LLMs for
- Assistants that answer questions over internal documents, combined with retrieval (RAG).
- Drafting and summarising emails, reports, contracts and meeting notes.
- Classifying and extracting data from unstructured text: tickets, claims, invoices.
- Customer service in chat and voice channels.
- Writing and reviewing code.
- Acting as the reasoning engine of AI agents that call tools and systems.
Limits to plan for
- It can produce fluent but wrong answers (hallucinations), so important outputs need sources or checks.
- Its knowledge stops at its training cut-off unless you give it current data.
- It has a finite context window: it can only consider so much text at once.
- Cost and latency grow with model size and with the length of prompts and answers.
- Sensitive data and regulation (GDPR, the EU AI Act) shape which model you can use and where it runs.
Frequently asked questions
What does LLM stand for?
LLM stands for large language model. "Large" refers to the number of parameters and the volume of training text; "language model" means a model that estimates which word or token is likely to come next.
Is ChatGPT an LLM?
ChatGPT is an application built on top of LLMs from OpenAI's GPT family. The LLM is the model; ChatGPT adds the chat interface, memory, tools and safety layers around it.
What is the difference between an LLM and generative AI?
Generative AI is the broader category of models that create content: text, images, audio, video or code. An LLM is the type of generative model specialised in language, and it is the most widely used one in business today.
How is an LLM trained?
First it is pre-trained on a very large text corpus to predict the next token. Then it is adapted with instruction tuning and, often, reinforcement learning from human feedback so that it follows requests and avoids harmful outputs. Companies rarely train one from scratch; they use an existing model and adapt it with prompts, retrieval or fine-tuning.
Can an LLM use my company's data?
Yes, usually without retraining it. The common pattern is retrieval-augmented generation: relevant documents are found and passed to the model with the question. Access permissions, data residency and logging need to be designed in from the start.
Which LLM is best for a business?
It depends on the task, the language, the cost per request, latency and where the data can be processed. Many organisations combine a large model for complex reasoning with smaller ones for high-volume, simple tasks, and test them on their own cases before choosing.
Decision guides that use this term
- Open-source vs proprietary LLMs: how should an enterprise choose?
- Single AI vendor vs multi-model strategy: should you bet on one provider?
- AI gateway vs direct API integrations: how should applications connect to models?
- Long context windows vs RAG: do you still need a retrieval pipeline if the model can read the whole dataset?
- Reasoning models vs standard LLMs: when is the extra thinking worth paying for?
- Seat-based vs usage-based AI pricing: which model fits how your organisation will use AI?
- Small vs large language models: which one should run your enterprise use case?
How to put it to work
Where it shows up at Thinkia
Further reading
Related terms
Sources
- Vaswani et al. · Attention is all you need (arXiv)
- Brown et al. · Language models are few-shot learners (arXiv)
- Ouyang et al. · Training language models to follow instructions with human feedback (arXiv)
Updated: 2026-10-21