Ask HN: Do you think language development will be frozen in time? | Hacker News Reader

Ask HN: Do you think language development will be frozen in time?

2 pointsparpfish5 comments

I was just thinking about how new languages slowly build adoption and had the thought that LLM based coding may have basically frozen things in time so new languages (or language features) won’t be adopted.

Without a large training corpus, agents/autocomplete assistants will struggle knowing what to do (or will just confidently do the wrong thing), so people will hesitate to adopt a new language even if it has great features.

Or maybe we’re going to see an explosion of new languages for every project because it’s one way to stay ahead of the machines and make sure that humans are writing your code?

View on HN

I program in Python, Fortran, and R. LLMs may be better at Python than Fortran since there is more training data for the former, but they are still pretty good at Fortran. LLMs make it easier to use less popular languages and much easier to use the more popular ones. An advantage of using LLMs for a compiled language like Fortran is that you can iterate with compiler messages until the code compiles. A Python script to do this is at https://github.com/Beliavsky/Groq-Fortran-agent. I assume Cursor and Windsurf do this too.

That’s a good point about compiled languages being able to bootstrap themselves a bit.

But in the case of a young language, I worry that LLMs will induce a sort of convergent evolution either in syntax or just idiomatic best practices

Language adoption maybe, but not language development.

Language dev requires precise semantics. If a user decides that types or other academic nonsense are overrated, they can get away with it. You can be cavalier and just let your application crash - as long as there's a precise formulation underneath you about what 'crashing' means: executing `finally` blocks, rolling back the stack and capturing a stack trace, following an exceptional continuation, etc.

You can't YOLO an Algorithm W, an ANF transformation, memory models or register allocation. Language design has to be just right. I don't think language design selects for people who look at a "95% correct" LLM benchmark and think "wow!". A program generated by a 95%-correct compiler will probably run catastrophically wrong in about 100% of executions.

I think we won't see new language development either explode or decline. The reason that there are so many different languages is that each one embodies different tradeoffs, and sometimes one set of tradeoffs fits the problem at hand better than a different set of tradeoffs.

I don't see any reason to think that this reality has changed or that LLMs will change things on this front.

    so people will hesitate to adopt a new language even if it has great features

Tools like LSPs etc are almost always behind the leading-edge of development; all that means is that the "red squiggles go brrrrrr" crowd doesn't adopt things right away.

Reply on news.ycombinator.com