Definition
A Large Language Model (LLM) is a neural network — typically a Transformer — trained on vast amounts of text to model the probability of the next token given previous tokens.
At inference time, LLMs generate text token by token. They can answer questions, summarize, translate, write code, and follow instructions when aligned with human feedback or prompting techniques.
In simple terms
An LLM is like an autocomplete engine that read the entire library of the internet. It doesn't 'know' facts the way a person does — it predicts the most plausible continuation of text based on patterns learned during training.
Where you see it
- ChatGPT, Claude, and Gemini are consumer-facing LLM products.
- GitHub Copilot uses LLMs for code completion.
- Customer support bots draft replies from company knowledge.
- Researchers use LLMs to analyze Somali text and build low-resource NLP tools.
How it works
1.Pre-training
The model learns general language patterns by predicting masked or next tokens on billions of words of text.
2.Tokenization
Input text is split into subword tokens; each token maps to a vector the model processes.
3.Autoregressive generation
The model outputs one token, appends it to context, and repeats until a stop condition.
4.Optional alignment
Fine-tuning or RLHF steers outputs toward helpful, safe, instruction-following behavior.
At a glance
Why it matters
- LLMs are the most accessible entry point to modern AI for developers and non-developers alike.
- They power retrieval, agents, coding tools, and translation — but only when you understand their limits.
Often confused
LLMs think like humans.
LLMs predict the next token based on learned statistical patterns — they do not have beliefs, memory, or consciousness.
LLMs always tell the truth.
They can hallucinate plausible-sounding but incorrect statements, especially on niche or recent facts.