Introduction to Ada: a project-based exploration with rosettas

Would the "LLM era" revitalize languages like Ada and Haskell into mainstream?

From the guidelines:

> Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something.

because... they don't have as many examples, documentation, textbooks, or public example projects to base generation off of, perhaps. There may be a future where documentation/servers are more formally integrated with LLMs/AI systems in a way that makes up for the relative lack of literature by plugging into a source of information that can be used to generate code/projects.

It's a not-so-ideal situation: how is the marketplace of libraries and languages going to evolve when you're competing against whatever version of Python and $FRAMEWORK that was crawled a long time ago?

If AI is writing the code, how important is it to have new languages?

That might actually be a benefit as most public code say in C++ is not good code.

If the pool is smaller but from say experienced programmers then the number of errors might be less. I can see that for Ada however most Haskell is probably written by undergraduates just learning it so not a quality code base.

I think Apple researchers published a recent papaer where they had a LLM giving good Swidt code but the original corpus only included one Swift program but the AI model was tuned by experienced Swift programmers to get into a good stae for general use.

I would say yes in that it could help revitalize things a bit. Writing difficult and complicated bindings to C libraries will be much easier now. Also, if you can supply a decent context, LLMs can do some good coding in Ada (just not new or fancy features without examples).

Claude does an okay job of translating from other languages into Ada. It works especially well if you write the specification (.ads) file and let it write the body (.adb)

Ada’s strictness about types and a preference to allocate on the stack rather than the heap means more bugs are caught at compile time. Claude Code is really good at iterating on compile time errors without much user intervention.

There was a pretty good article a while ago on how using verified SPARK (a subset of Ada) could help with llm generated output https://arxiv.org/html/2502.07728v1

LLM will make all languages mostly irrelevant, a niche like Assembly programming is nowadays, it is the next abstraction level, generating existing languages is only a temporary measure until they get good enough to generate straight executables.

Not even the deterministic high-level programming languages have succeeded to make the assembly languages irrelevant, despite everybody claiming that this will happen.

While the amount of source code written in assembly languages is an extremely small fraction of the total existing code and only few programmers are competent to write such programs, that assembly source code determines a large fraction of the performance of the applications run on modern computers.

LLMs are likely to behave similarly, i.e. a good amount of programs will continue to be written directly in deterministic programming languages by competent programmers, while a greater amount of source code, usable for solving problems that are neither novel nor critical, will be generated by people with lower skills, with the help of LLMs.

I predict that it will never make sense for artificial neural networks to directly generate machine code, for most of the same reasons it doesn't make sense for biological neural networks to do so.