A large language model (LLM) is a type of artificial intelligence model trained on massive volumes of text to understand and generate human language. LLMs are the technology behind AI assistants and chatbots such as ChatGPT, Claude and Gemini. The “large” refers both to the size of the training data (much of the public internet, books and code) and to the number of parameters – the internal weights the model adjusts during training – which can run into the hundreds of billions or trillions.
Almost all modern LLMs are built on the transformer architecture. At its core, an LLM does one thing: given a sequence of text, it predicts the most probable next token (a word or word fragment). Repeated over and over, this simple mechanism produces fluent essays, working code, translations and step-by-step reasoning.
Building an LLM typically involves three stages:
- Pre-training – The model learns language patterns by predicting missing or next tokens across an enormous general-purpose text corpus.
- Fine-tuning – The pre-trained model is further trained on smaller, curated datasets to specialize it for tasks such as following instructions, coding or medical Q&A.
- Alignment – Techniques such as reinforcement learning from human feedback (RLHF) teach the model to be helpful, safe and honest by learning from human preference ratings.
LLM capabilities include text generation, summarization, translation, sentiment analysis, question answering, code generation and acting as the reasoning engine inside AI agents. Well-known model families include OpenAI’s GPT series, Anthropic’s Claude, Google’s Gemini and Meta’s open-weights Llama models.
LLMs also have important limitations. They can hallucinate false information, their knowledge is frozen at a training cutoff date unless connected to external tools or search, they can reproduce biases present in training data, and they are limited by a context window – the maximum amount of text they can consider at once.
