# Large Language Model (LLM)

AI Glossary · LLMs & Generative AI
Canonical URL: https://aisat.uk/glossary/large-language-model
Author: Lucy Gilmore, Assistant General Counsel, Product, Privacy & AI at Sprinklr | Practical Privacy, Product & AI Governance & Defensible Risk Decisions (https://www.linkedin.com/in/lucy-gilmore/)
Breadcrumb: [Glossary](https://aisat.uk/glossary) > [LLMs & GenAI](https://aisat.uk/glossary?category=llms) > Large Language Model (LLM)

## TL;DR

> **A Large Language Model (LLM)** is a deep neural network built on the transformer architecture and trained on trillions of tokens of text and code to predict the next token in a sequence. At scale, this objective yields broad capabilities in reasoning, translation, coding, and synthesis.

## How Large Language Model (LLM) works in practice

An LLM processes input by first breaking raw text, code, or multimodal inputs into sub-word units called tokens via a tokeniser. Each token is mapped to a high-dimensional vector and passed through dozens of stacked transformer layers. Inside each layer, the self-attention mechanism computes how strongly every token in the context window relates to every other token, allowing the model to resolve long-range dependencies across thousands of words.

Building a production LLM happens in three phases. First, self-supervised pre-training on massive web, book, and code corpora builds a base model that predicts the next token statistically. Second, Supervised Fine-Tuning (SFT) on curated instruction-response pairs teaches the model to act as a helpful assistant. Third, post-training alignment using Reinforcement Learning from Human Feedback (RLHF), Direct Preference Optimisation (DPO), or Reinforcement Learning with Verifiable Rewards (RLVR) refines safety, instruction adherence, and reasoning accuracy.

When a UK retail bank uses an LLM to draft mortgage underwriting summaries from broker notes, the model is not consulting a built-in database of facts; it is computing probability distributions over tokens conditioned on the documents placed in its context window.

## Why it matters and trade-offs

LLMs provide a universal natural-language interface across unstructured enterprise work: code generation, contract analysis, customer support, and data extraction can all be prototyped in hours using the same foundational API. Both proprietary frontier models (Claude, GPT, Gemini) and open-weight models (Llama, Qwen, Mistral) now support structured JSON outputs, tool calling, and multimodal vision.

Because an LLM is fundamentally a probabilistic next-token predictor rather than a deterministic database or calculator, it will confidently fabricate plausible-sounding facts when asked about information outside its context window. Production engineering therefore centres on surrounding the LLM with retrieval (RAG), deterministic tools, guardrails, and automated evaluations rather than trusting raw model outputs alone.

## LinkedIn explainer post

- [Lucy Gilmore on Large Language Model (LLM) (LLMs)](https://www.linkedin.com/feed/update/urn:li:share:7524053868075466752/)

## Related AI concepts

- [Small Language Model (SLM)](https://aisat.uk/glossary/small-language-model): A Small Language Model (SLM) is a compact neural language model, typically between one billion and ten billion parameters, optimised for low-latency, low-cost inference on a single GPU or on-device NPU for domain-specific workflows.
- [Reasoning Model](https://aisat.uk/glossary/reasoning-model): A Reasoning Model is a large language model trained with reinforcement learning to expend additional inference-time compute generating, critiquing and verifying multi-step internal thought chains before emitting its final response.
- [Context Engineering](https://aisat.uk/glossary/context-engineering): Context Engineering is the engineering discipline of dynamically selecting, structuring, compressing and ordering the system instructions, retrieved knowledge, tool definitions, memory and examples placed inside a language model's context window at inference time.
- [Hallucination](https://aisat.uk/glossary/hallucination): Hallucination is a failure mode in which a generative AI model outputs confident, fluent statements that are factually false, fabricated or unsupported by source documents because it predicts statistically plausible tokens rather than verifying truth.