Knowledge MaxxingLarge language model
A large language model is a neural network trained on vast amounts of text to predict and generate language, with enough parameters to capture broad patterns in grammar, facts and style.
Prerequisite: Deep learning
What matters
Scale in both data and parameters is what lets these models handle a wide range of language tasks with one general-purpose model, instead of needing a separate model trained per task.
Example
The same large language model can draft an email, answer a trivia question, and write code, without being retrained between tasks.
Common misconception
A large language model does not 'understand' language the way a human does; it predicts plausible continuations based on learned statistical patterns.
Recall question
What lets one large language model handle many different language tasks without retraining?