I wonder if it's possible to train to read text encoded as one colored pixel per letter, or even per token.

I can't really read anything with that, so somewhat readable is very moot.

I love this. It speaks to me in a similar ways as a lot of the AI zeitgeist—why shouldn’t we optimize for how the brain actually operates at scale versus hundreds-years-old ideas about ligatures designed for reading in candlelight? (In the AI case, a romanticism for having to learn and prove memory in such a rote way)

Pity there's no italics ...

SCNR

A thread last year with lots of related subpixel type things:

Nanofont3x4: Smallest readable 3x4 font with lowercase (2015)

https://news.ycombinator.com/item?id=39735675

okay but what about "c" being nearly the same as "z", neither of which look like the character and are nearly(?) identical. Is our brain supposed to just be able to figure it out?

The Atari 2600 had pretty good vertical resolution (assuming you could set up the next line in 76 cycles) but limited horizontal resolution. A 3x5 font is possible, but good luck distinguishing N from M.

This font seems to use characters up to 5 pixels wide, which helps with its near-legibility.

Is it just me or the s Z and z S should be swapped?

Love this. Brings so much joy. Try some punctuation. Hilarity ensues.

Cool. I hate it.