This little program is learning the alphabet the way a curious child might — by
seeing shapes, hearing sounds, and noticing that they belong together. It runs right
here in your browser. Press a letter and it will say the sound and draw the shape; press a
sound and it draws the letter; or draw your own and watch it guess. First time here?
Start on the home page →
Honest banner. A signal-prior simulation — it does not see, hear,
understand, or read. Binding is co-occurrence counting. The audio you hear is a lossy synthesized
tone (a demo), not speech. Citable numbers come from pytest
/ challenge, not this page.
checking JS≡Python…
Show it a letter → it says & draws
picture → cause → read-out → speak + draw
shown
→
·says (▶)
read-out
→
it draws
Play it a sound → it makes the picture
sound → cause → read-out → draw
🔊sound
heard
→
it makes
→
·says (▶)
and says
Draw your own glyph
the honest boundary: it often misreads hand-drawing — and now shows you a real “match quality” so you can see when it is only guessing
UNI Needs some Glasses?A/B its eye
Same math, sharper eye. The 16×16 model has 256 receptors (4× the detail of 8×8's 64) so Devanagari and Latin letters come through crisp. Click 8×8 to see the degraded reference (that's the version whose parity is checked against Python; see the badge above). Nothing else changes — the model, the counting, the honesty banner all stay identical.
your signal (16×16)
Draw a letter (try A, B, G, K, M) — big and centered — then press Test it.
Expect it to miss a lot. The eye can be a crude 8×8 blur or a sharper 16×16, using the Needs some Glasses? toggle above,
not OCR — so hand-drawing often lands on the wrong letter. That is not the interesting part. The
interesting part is that it reports an honest match quality: when your drawing matches nothing well,
it says so, instead of pretending to be 100% sure. That is what honest confidence looks like.
UNI.Sensorium v0 · in-browser port ≡ Python reference (parity above) · no LLM, no backprop ·
real audio/graphics are non-citable demos.