Explainer

Next-Token Prediction

The task every language model is trained on, in one screen.

Placeholder explainer prose goes here, above the widget.

The cat sat on the

mat 18.7%

floor 14.2%

ground 9.8%

chair 7.8%

couch 6.2%

bed 5.1%

table 4.5%

windowsill 3.2%

step 2.7%

counter 2.4%

+ 151,926 other tokens with vanishingly small probability.

Optionally more prose after the widget.