Large Language Model (LLM)
A Large Language Model (LLM) is a deep neural network — almost always a transformer — trained on hundreds of billions to trillions of words to predict the next token, and to generate, summarize, translate, or reason over text.
Full definition
LLMs scale three things together: parameters (often 7B-2T+), training tokens (1T-15T+), and compute (measured in FLOPs). The "large" threshold is loose — modern frontier LLMs are typically 70B+ parameters. Their capabilities emerge from scale and self-supervised pretraining, then are shaped by post-training (instruction tuning, RLHF, RLAIF) into chat assistants. Examples include GPT-4o, Claude 4.5, Gemini 1.5, and Llama 3.
Why it matters
LLMs are the first general-purpose AI systems most enterprises adopt. McKinsey's 2024 State of AI report found that 65% of organizations now regularly use generative AI in at least one function — almost double the prior year. Choosing which LLM to standardize on is now a multi-year platform decision affecting cost, governance, and switching cost.
Example
A law firm uses Claude to summarize a 600-page deposition into a 4-page brief in 90 seconds; the same model also extracts citations, translates exhibits, and drafts opposing counsel responses.
Frequently asked questions
What does the "L" in LLM mean exactly?
"Large" historically meant billions of parameters trained on hundreds of billions of tokens. There is no formal cutoff, but production LLMs today are typically 7B-2T+ parameters.
Are LLMs the same as ChatGPT?
No. ChatGPT is a product built on top of an LLM (originally GPT-3.5, now GPT-4o and successors). The LLM is the underlying model; the chat interface, system prompts, tools, and safety layers are added on top.
Related terms
- Foundation ModelA foundation model is a large model trained on broad data at scale — typically self-supervised — that can be adapted to many downstream tasks via prompting, fine-tuning, or retrieval, instead of being trained task-by-task.
- TransformerThe transformer is a neural-network architecture built around the self-attention mechanism that has become the dominant model design for language, vision, audio, and multimodal AI since 2017.
- Token / TokenizationA token is the basic unit a language model reads or writes — usually a sub-word fragment (about 4 characters of English text) — and the unit by which API pricing, context-window limits, and inference cost are all measured.
- Context WindowA context window is the maximum amount of text — measured in tokens — a language model can read and reason over in one inference call, equivalent to the model's working memory for that turn.
- Prompt EngineeringPrompt engineering is the practice of designing the inputs to a language model — instructions, examples, role definitions, output formats, and constraints — to reliably produce a desired output without changing the model's weights.
- Retrieval-Augmented Generation (RAG)Retrieval-Augmented Generation (RAG) is a technique where a language model retrieves relevant documents from an external knowledge base before generating a response, reducing hallucination by 30-60% and enabling citation of sources.
Source & further reading
Primary source: Brown et al. — "Language Models are Few-Shot Learners" (GPT-3 paper) (2020).
Citation policy: this entry is part of the AIDOLS AI Implementation Glossary and may be quoted for research, journalism, and education with attribution to aidolsgroup.com/it/glossary/large-language-model/.