no button · reacts to selection · runs on your GPU
The model is reading, not writing.
Select any stretch of the text below. Every character is tinted by how surprised the model was to find it there — one forward pass over your whole selection, on your own GPU.
0 – 6 bits · fixed scale, not normalised per selection
Your selection, read on its own
What you are looking at
Before each character, the model has read everything before it and formed an opinion about what comes next. The tint is the probability it gave the character that actually turned up, in bits: 0 means it was certain, 6 means it was as good as guessing.
Why it is fast
Scoring text you already have is not generation. Every position is predicted from the text to its left, and all of them can be done at once — so a paragraph costs a single forward pass, not one step per character. The model is loaded once, when the page opens.
Why it is a better demo
Every other demo asks you to compose a prompt, which is work. This one asks nothing: it reacts to reading. And a model that reacts to a selection in a few milliseconds is a model that is genuinely on your machine — there is no round trip that could be that fast.
The model here is the 10.6 M-parameter character-level Shakespeare model that ships in the Forge repository. It knows 65 characters and nothing else, so it finds ordinary modern prose genuinely surprising — that is a real result, not a bug.