The little program on our other pages has a very tiny eye — only 64 little
dots sampling each letter. That's why its drawings look blocky. Here you can give it a
bigger eye. Draw a letter below, and three eyes — a tiny one, a medium one,
and a big one — will each sample your drawing, store it, and draw back what they
stored. You'll see the little eye squish everything into a blob, and the big eye keep the shape.
That's why eyes matter.
A tiny simulation, not a real eye. Your drawing stays in your browser —
nothing is uploaded or saved. "Learning" here just means the program is keeping tally: which
shapes and which labels tend to appear together.
How to play: draw a letter in the pad on the left, pick the letter you drew from the
dropdown, then press learn this drawing — the three eyes will each store it. Press
recognize and each eye will guess what you drew. Press basic training to give them a
quick round of practice on all five built-in letters (A, B, G, K, M). Additive means what
they've stored stays; from blank wipes everything they stored so they hold only what YOU
teach.
Draw here
click / touch & drag
label
draw a letter, pick its label, then “learn”.
training
Each eye shrinks your drawing down to fit its own grid, so the 8×8 stored prototype
looks blocky and the 32×32 one keeps the strokes. Additive means new drawings pile on
top of what it already stored — watch a letter's stored prototype get sharper as you draw more of it.
From blank wipes everything it stored first, so it holds only what YOU teach.
UNI·Sensorium Vision Lab · vision-only, no audio · same InferenceCell/CountTable math as the being · no LLM, no backprop.