Behind an AI response
A text-generating model learns patterns during training. At inference time, it processes tokens and context, estimates possible next tokens, and repeats selection to create a response. It can still generate inaccuracies.
Visual explanation
01
Tokenize
Text becomes smaller model tokens.
02
Process
The network uses context and attention.
03
Predict
A continuation token is selected.
04
Repeat
Tokens accumulate into a response.
