Claude and the like are a huge problem for new languages that want to do new things. It was bad enough when a LaTeX replacement had to compete with forty-ish years of package development time. Now they also have to compete with the millions of lines of existing code LLMs have hoovered up.
I've done some simple Typst programming via Claude, and it worked fine. I expected it to be ignorant of Typst but that was not the case.
One of the best things about Typst is that most tasks are very simple. Compared to the reams of Latex BS I was replacing, building my book with Typst is momumentally simpler.
Sure but if the LLMs are making LaTeX easy to work with then why bother trying migrate everyone to a new language?
I think there are a bunch of assumptions behind your statement that I believe are not true:
1. Latex is sufficient for all document publishing needs. E.g. converting Latex to HTML is bad to non-existent, while Typst has HTML export.
2. LLMs are sufficient for solving all problems one can encounter.
3. Things that are easier for humans are not also easier for LLMs.
4. New releases of LLMs will not learn more about Typst
At the end of the day I'm not trying to migrate anyone. Use whatever you feel is best. For my use cases I'm convinced Typst is a better option than Latex.
> why bother trying to migrate everyone to a new language?
.. because a new language might be better?
But moving forward it’ll be harder to tell if any given new language is better than existing alternatives. LLMs burden their users with an almost insurmountable status quo bias.
Because LaTeX is ugly to write and not human-friendly. Adding an AI agent to the loop does not fix those issues.
Great for code re-use but I agree, terrible for anything new.
Which is good, because we don't want to deal with inferior solutions to typesetting that pop up every few years.
A slight bias in favor of the status quo might be acceptable or even desired. However current LLMs strongly favor traditional languages and are unable to comprehend even modern language features not part of their base training set.
Consider the counterfactual of LLMs being available in the 1990s, trained mainly on the world's C code. Perhaps we would still be exclusively writing C today for new languages' code could not been synthesized as easily or conveniently. It's not just about Typst or typesetting specifically but programming language design in general and that improvements are becoming much harder to push through.
> Perhaps we would still be exclusively writing C today for new languages' code could not been synthesized as easily or conveniently.
I'm not actually sure that would be a bad thing? All the reasons that immediately come to mind to move away from C have to do with ergonomics and safety, the latter largely being a product of the former IMO. If an LLM can ingest my entire codebase and do 90% of the work to get me to the changes I need doesn't that obviate the majority of the motivation to change languages in the first place?
If we get a completely autonomous AI it probably won't program in C (or any other human-understandable language).
If it still programs in a human readable language, that means people need to review the code, at least from times to times. And it's much easier to review modern languages than C.
Have you tried typst at least once. You have big words but it is lightyears better than Latex.