Although this feels so LLM-generated I can't bring myself to read more than a quarter of it, I have been wondering about the general premise for quite some time now.

  • New language bad when humans need to learn to write it
  • AI overload bad, too much to review
  • Make new thing easy to read, doesn't matter if it's hard to write because AI can do that

I don't do much in 3d (only shadertoy, and light CAD for 3d printing), but I do sometimes see videos from game artists, and see node-based editors. That seems useful, how hard is a diff in node-based stuff? https://en.wikipedia.org/wiki/File:Shader_nodes_in_blender.p...

I remember UML from my university course (2002-2006), though I think I only ever used it once, in my industrial year during that degree. Again, how hard is review? I don't recall code review or version control being part of the degree, it certainly wasn't part of the UML course.

Even a decade or so ago, I was thinking "gee, why can't we compile a LaΤεΧ equation directly as a function? Render it in the editor as it would appear in arxiv?"… but absolutely couldn't be arsed to write a parser that would do that. Now LLMs are good enough I can do this for a joke language written entirely in Norse runes*, so this may be worth experimenting with? But I say "may" not "is" because there's so much pressure to just "accept all" from the AI without looking that I don't know if even this would make any meaningful dent in code quality or worker stress reduction.

* Under no circumstances should this be taken seriously: https://github.com/BenWheatley/RealFutharkLanguage