AI glossary / How it works

Test-time compute

How it works, also called Inference-time scaling

Letting a model think for longer on a hard question to get a better answer, at the cost of speed and money.

More words in How it works

• TokenThe small chunks of text an AI reads and writes in, roughly three-quarters of a word each. Limits and prices are usually counted in tokens.
• TokenizationChopping text into tokens before a model reads it.
• TrainingThe learning stage, where a model is shown vast numbers of examples and adjusts itself until it gets good at the task.
• Training dataThe material a model learned from. The quality of what goes in shapes the quality of what comes out.
• Transfer learningReusing what a model learned on one task as a head start on another.