A Large Language Model (LLM) is a neural network trained on large corpora of text using self-supervised objectives — most commonly next-token prediction — to learn statistical relationships between words, phrases, and concepts across an enormous vocabulary. These models use the Transformer architecture and typically contain billions of Parameters, enabling them to perform a wide range of language tasks — summarisation, translation, code generation, question answering, and reasoning — without task-specific retraining. Their capabilities emerge from scale: both the size of the model and the volume and diversity of the training data.

Think of it like a system that has absorbed an enormous fraction of written human knowledge and learned, at a very fine-grained level, how ideas connect and how sentences are typically constructed — not by memorising text verbatim, but by internalising patterns well enough to generate coherent new text in any style or domain.

A concrete illustration is chain-of-thought reasoning. Given the prompt "If a train travels 60 km/h for 2.5 hours, how far does it go?", a modern LLM does not just retrieve an answer; it generates intermediate steps:

Distance = speed x time
         = 60 km/h x 2.5 h
         = 150 km

This step-by-step generation emerges from Pretraining on text that includes worked examples, and can be further improved through RLHF and Fine-tuning on curated instruction data.

LLMs sit at the intersection of many foundational concepts: Tokenization determines how text is fed in, Attention governs how the model relates tokens to one another, Context Window limits how much text can be processed at once, and Hallucination describes the failure mode where the model generates plausible-sounding but incorrect content. LLMs are the core component inside Foundation Model systems and are frequently deployed via techniques like LoRA and Quantization to make them practical on constrained hardware.

You will encounter LLMs in virtually every modern AI product that processes or produces natural language, from chatbots and coding assistants to document summarisers and enterprise search systems.