Skip to content
Nitmonk
Language Models

What is a Large Language Model (LLM)?

Inside an LLM — tokenization, embeddings, transformer.

What is a Large Language Model (LLM)? — infographic explaining Inside an LLM — tokenization, embeddings, transformer.
What is a Large Language Model (LLM)? — visual explainer by Nitmonk.

A Large Language Model (LLM) is an AI model trained on trillions of words to understand and generate human-like language. Inside, your prompt is turned into tokens, then numbers (embeddings), processed by a transformer, and turned back into a coherent answer.

In simple terms

You give a prompt; the model tokenizes it, processes it through a transformer, and predicts the next tokens.

How it works

  1. 1Input: you give the model a prompt or question.
  2. 2Tokenization: the text is broken into smaller pieces called tokens.
  3. 3Embedding: tokens are converted into numbers the model can understand.
  4. 4Transformer: the model analyses patterns and relationships across all tokens.
  5. 5Output generation: it predicts the next tokens, one by one.
  6. 6Response: you get a coherent, context-aware answer.

Key points

  • Understand: it grasps meaning, context, instructions and intent.
  • Generate: it writes, answers, summarises and explains.
  • Reason: it can solve problems and reason step by step.
  • Generalise: it applies knowledge to new topics it hasn't seen before.

Why it matters

LLMs are the foundation of modern AI applications — chatbots, copilots, search, coding assistants and research tools. Knowing the internal pipeline helps you prompt them better and understand their limits.

Frequently asked questions

What is a token?
A subword unit of text. Models read and generate tokens, and both cost and context limits are measured in tokens.
What are examples of LLMs?
GPT-4o, Claude, Gemini and Llama are all large language models.

More in Language Models

Large Language Model (LLM) — The Basics

Massive data → training → understanding → generation.

GPT (Generative Pre-trained Transformer)

How GPT predicts text, one token at a time.

GPT-4

OpenAI's multimodal model — text and images.

ChatGPT

From your question to the answer, step by step.

Claude

Anthropic's helpful, harmless and honest AI assistant.

PaLM / Bard

Google's conversational AI powered by PaLM.

Back to all AI concepts