[superconduct123]

> more information compression (see paper) => shorter context windows, more efficiency

I'll ask the dumb question here

How is that possible? Wouldn't different sizes of text/fonts/rendering/spacing end up with way worse compression?