What is a large language model (LLM)?

A large language model, or LLM, is an AI system trained on enormous amounts of text. Its core skill sounds simple: predict the next word. Done at massive scale, that skill turns into writing, summarizing, translating, coding and answering questions.

How it learns

• Pretraining: the model reads vast amounts of text and learns by guessing the next piece of each sentence, over and over. No one labels the data; the text itself is the lesson (this is called self-supervised learning).
• Tokens: computers work with numbers, so text is first split into small chunks called tokens and turned into numbers.
• Fine-tuning: after pretraining, models are refined to follow instructions and behave helpfully. One common method, reinforcement learning from human feedback (RLHF), trains the model toward answers people prefer.

The transformer

Most LLMs use the transformer, a neural network design introduced by Google researchers in the 2017 paper "Attention Is All You Need". Its "attention" mechanism lets the model weigh how every word in a passage relates to every other word at once.

Where you meet them

Chatbots such as ChatGPT, Claude, Gemini, Grok and DeepSeek are built on LLMs. Model families include OpenAI's GPT series, Meta's Llama and Mistral's models. Increasingly, LLMs also power AI agents that take actions, not just write text.

Limits to know

• Hallucinations: LLMs can state false things confidently, so important facts need checking.
• Bias: models reflect the data they learned from, including its gaps and biases.