>Why don't you make them _S_ shaped?

To some degree, this problem was eventually solved, c.f., the five volume set _Computers and Typesetting_:

https://www-cs-faculty.stanford.edu/~knuth/abcde.html

but then one had the effort to create a new typeface set for math equations by the AMS, eventually named Euler as written up in "AMS Euler — a new typeface for mathematics". _Scholarly Publishing_ and so forth, but arguably, things went awry in that rather than capture the ductus of Prof. Zapf's pen, and model based on that stroke and a pen shape, the expedient approach of simply modeling the outline was arrived at and implemented due to the difficulty and lengthy time required for the idealized approach.

Another consideration may have been that there doesn't seem to be an available algorithm which is robust and accurate and automatic for determining the curves which describe the union of arbitrary Bézier curves (some projects get around this by making high resolution pixel images and tracing them).

Fig 9 stood out to me as obviously wrong. The two glyphs on the left are pixel by pixel identical, as are the three middle ones, and the two on the right. Quite mysterious though considering this PDF appears to be a scan.

See the definitions of "O" and related glyphs for a good example[2].

[1] https://github.com/be5invis/PatEL

[2] https://github.com/be5invis/Iosevka/blob/main/packages/font-...

Having models attempt an SVG letter S remains one of my personal/informal LLM benchmarks. They are still pretty bad at it.