noun · AI concept
LLM
also called large language model
A very well-read autocomplete. It has seen so much writing that it can continue almost any text in a believable way.
A large language model: a neural network trained on large amounts of text that generates text by predicting the next token.
A neural network with billions of parameters, trained to predict the next token over a very large body of text. Chat, code and summaries all come from that one skill, applied one token at a time.
Real-life example
You ask a chat tool to turn a messy list of notes into an email. It does it well, so it feels like it understands your job. What it did was continue your notes in the shape of an email, because it has seen thousands of emails. Ask for something rare and the same skill produces something confident that fits less well.
Where the analogy breaks
-
It is not a search engine or a database. It does not look up stored answers, it generates text that fits the input.
-
A fluent answer is not a checked answer. The model writes what sounds right, so it needs a source or a test when facts matter.
-
The LLM is not the whole product. Chat apps add memory, tools and safety around it, so the same LLM can feel different in different apps.