How it learns: guess, check, nudge

Each step raises the real next word’s probability — loss falls.

“The cat sat ___

Loss
0.00
−log( P of “on” )
Loss over steps
step 0

The bar for “on” (the true word) rises; the rest fall and renormalize.