Glossary — Language models
Training
Training is the process where a model looks at an enormous number of examples and adjusts its internal numbers until it predicts better.
1 tokensTraining
Simply put#
A model starts with random numbers and is completely useless. It is shown text, predicts the next token, gets it wrong, and the numbers are nudged slightly. This repeats trillions of times.
Example#
Training a large language model takes months and costs tens of millions. That is precisely why there are so few large models and why they belong to large companies.
Why it matters to you#
Training finished before you ever used the model. It does not "learn" from your conversation — what you tell it lives only in the context window and disappears when the conversation ends.