Glossary
Large language model (LLM): what it is
A large language model (LLM) is an AI model trained on very large amounts of text to predict and generate language, which lets it read, summarise, write, translate and reason through problems. Most modern LLMs are built on the transformer architecture introduced in 2017. In business, an LLM is the engine inside an agent, not the whole system.
Featuring insights fromShaun Devan · Founder & CEOUpdated · 3 min read
The model vs the system around it
| The LLM on its own | An agent built on an LLM | |
|---|---|---|
| Knows | What it learned in training | Your data, through retrieval and tools |
| Can act | No, it only produces text | Yes, through tools and system access |
| Is tested on | General benchmarks | Your own past cases |
| Limits | Whatever the user asks | A written authority line |
Choosing a model
Different steps of one workflow often use different models: a cheaper, faster one to classify documents, a stronger one for judgement calls. Because models change every few months, the system should be built so a model can be swapped and re-tested against the evals without rebuilding everything else.
Common limits
- It can state wrong facts fluently, so answers need grounding and checks.
- It does not know your data unless you give it access.
- Its output varies run to run, which is why evals matter.
Sources
Frequently asked questions
What is an LLM in simple terms?
A large language model is an AI trained on huge amounts of text so it can understand and produce language: answer questions, draft text, summarise documents and reason through tasks.
What is the difference between an LLM and an AI agent?
An LLM generates text. An AI agent uses an LLM to decide what to do, then acts through tools and systems to complete a task.
Which LLM is best for business use?
It depends on the task. Many production systems use several models, matched to each step on accuracy, speed and cost, and tested on the business's own cases.
