AI Trend Research

← AI glossary

AI glossary · Beginner · Language models

LLM

An LLM (large language model) is an AI program that learns from a very large amount of text so it can understand and write text the way people do. It answers questions, summarizes, translates and writes.

About this term
Level
Beginner · Core ideas for first-time readers
Category
Basics
Added
2026-10-07
In plain words It is like a 'writing master' who has read countless books and has a feel for how sentences flow. The difference is that it does not know the world through experience as people do; it has learned patterns in text.

How it works

An LLM is trained to look at small pieces of text (tokens) and predict the next piece. Chaining these predictions together produces long answers. Training uses an enormous amount of text, and in the process countless numbers inside the model (parameters) are adjusted. Most LLMs today are built on an architecture called the 'Transformer.'

Adding the next token one at a timeInput → Split into tokens; Split into tokens → LLM; LLM → Next-token candidates; Next-token candidates → Pick one; Pick one → Append to input; Append to input → LLM(Repeat)Input'Today the weather is'Split into tokensLLMTransformer ·countless parametersNext-tokencandidates'sunny' · 'cold' · …Append to input'Today the weather issunny'Pick oneRepeat
Adding the next token one at a time An LLM looks at the text so far, picks one next token, appends it to the input and predicts again. This repetition produces long answers.

Why it matters

LLMs are the core engine of many AI services in use today, including chatbots, search, coding assistants and document writing tools. A big change is that they can handle many tasks just from instructions written in plain language, without building a separate program for each task. That is why companies and countries are competing to build or adopt their own LLMs.

Things to know

LLMs can produce content that sounds plausible but is wrong, which is called 'hallucination.' They also do not know about information from after their training cutoff unless it is provided to them. It is a good habit to check sources for important information. It also helps to remember that each model is better at different languages and domains.

Example

Some LLMs, like 'Falcon-Emirati' featured on this site, are built to understand the dialect and culture of a specific region. It is an example of a model trained mainly on text from that region so it handles local expressions and customs that general-purpose LLMs tend to miss.

Where it came up on this site

Related terms

Token Parameter Transformer Hallucination Prompt Fine-tuning