What is it?
An LLM is a model trained to work with language tokens.
It learns statistical patterns in large text datasets and uses those patterns to predict likely next tokens. A product such as ChatGPT may use an LLM plus many other systems.
02 · LLMs & Transformers
What is an LLM?
An LLM is a model trained to work with language tokens.
Think of it like...
Think of it like an autocomplete engine trained on a huge library. It can continue a request in useful ways, but it does not retrieve a guaranteed fact list from a database.
How it works
An LLM repeatedly predicts the next token until it reaches a stopping condition.
1 · Input tokensReceive textThe prompt is split into tokens.
2 · Transformer layersProcess contextThe model combines information from the current sequence.
3 · Next-token scoresEstimate optionsIt assigns probabilities to possible next tokens.
4 · RepeatExtend the answerThe selected token joins the sequence and the loop continues.
5 · OutputReturn textThe token sequence is decoded into readable text.
Real-world examples
Everyday example · Autocomplete
Input: The start of a sentence
System: The LLM scores possible continuations
Output: A suggested completion
AI product example · Chat assistant
Input: A user question
System: The product sends context to the language model
Output: A generated reply
Business example · Draft summary
Input: A meeting transcript
System: The LLM follows a summary instruction
Output: A draft summary for review
What it is NOT
LLM ≠ ChatGPT
LLM: A type of language model
ChatGPT: A product built with models, policies, interface, and operations
LLM ≠ Artificial Intelligence
LLM: One model type for language
Artificial Intelligence: The broader field
LLM ≠ Search engine
LLM: Generates text from learned patterns
Search engine: Retrieves existing information
Related concepts
Text → Tokens → Transformer → Next-token loop → Text
An LLM is a model component. An AI product may add retrieval, tools, memory, and safety controls.
Explore next
Remember this
An LLM generates language by repeatedly predicting the next token from context.
Independent video · Azure Neural voice