What happens if you present an image of a page in some font to an LLM, and ask it to make you a font file for that font? An LLM could probably not only do that, but create matching characters for ones not already present. Oh, and tell it to fix the kerning.
As a matter of US law: - Font letterforms are not copyrightable. (Eltra Corp. v. Ringer, 579 F.2d 294 (4th Cir. 1978) https://en.wikipedia.org/wiki/Eltra_Corp._v._Ringer >) - Font programmes are. (Generally 1980 Computer Software Copyright Act.) - Cleanroom reimplementation of software is not copyright infringement. An AI without access to original source code would likely pass this test. - As a further twist, it app…
A good question to work on is whether an LLM, given a reasonable number of characters of a font, can generate letter forms for the entire Unicode character space. Probably. It's a style matching problem, and LLMs are good at that.