What it means in practice
LLM stands for Large Language Model. The simplest way to picture it is as the engine, or the brain, of an AI.
Many people confuse ChatGPT with the model. ChatGPT is in fact only the chat application, the bodywork you log into. The actual brain generating the answers, the engine under the bonnet, is the LLM: a model called GPT-4o, for instance. The competing application Claude uses the Claude 3.5 Sonnet model.
These models have read an enormous slice of the internet. That is how they picked up the rules of language, grammar, facts and logic, and why they can hold a fluent conversation in languages they were never explicitly built for.
Why companies need to know this
Business owners do not need to understand the maths behind LLMs, but they do need to know that there is not just one model, and they are not all the same.
- Cost and performance. There are large, very capable models (GPT-4o, Claude 3.5 Sonnet) that are more expensive and slower. And there are smaller, less capable ones (GPT-4o-mini) that are extremely fast and cost a fraction as much.
- Choosing the right tool. If you want AI to analyse a 50-page legal contract, you need the biggest and smartest model. If you only want it to sort 500 incoming emails into “complaint” and “order”, a small cheap model is enough.
- Security and privacy. Some models, such as Meta’s Llama, are open source. You can download them and run them on your own servers, so your sensitive data never leaves the company.
An example from practice
A common mistake when automating company processes is using the most expensive model for every small task. It is like driving a lorry to buy bread rolls.
In our company AI workshops we teach teams how to pick the right LLM for a specific job. Choosing the right engine can save hundreds of euros a month in running costs, with the quality of the result completely unchanged.