FRESH

Hacker News

LLMs as Compilers

34 points by kadhirvelm

by ryanobjc

2 subcomments

So this has already been conceived of many decades ago, and there are some substantial issues with it, the illustrious djikstra covers it: https://www.cs.utexas.edu/~EWD/transcriptions/EWD06xx/EWD667...
Now this isn’t to say the current programming languages are good, they are generally not. They don’t offer good abstraction powers, typically. You pay for this in extra lines of code.
But having to restate everything in English, then hoping that the LLM will fill in enough details, then iterating until you can close the gaps, well it doesn’t seem super efficient to me. You either cede control to what the LLM guesses or you spend a lot of natural language.
Certainly in a language with great abstractions you’d be fine already.

by daxfohl

1 subcomments

Nah, I think that's the opposite of what to do. That requires you to specify all requirements up front, then press go and pray. Even if it worked perfectly, it takes us back to the stone ages of waterfall design. With LLMs, missing one requirement that would be obvious to a human (don't randomly delete accounts) often leads to a fun shortcut from the LLM perspective (hey if there's a race condition then I can fix it by deleting the account)!
The real value of LLMs is their conversational ability. Try something, iterate, try something else, iterate again, have it patch a bug you see, ask if it has recommendations based on where you are headed, flesh things out and fine tune them real time. Understand its misunderstandings and help it grasp the bigger picture.
Then at the end of the session, you'll have working code AND a detailed requirements document as an output. The doc will discuss the alternatives you tried along the way, and why you ended up where you did.
It's much like this in graphics too. Yeah you could spend a ton of time coming up with the single one-shot prompt that gives you something reasonably close to what you need, which is how it worked in the past. But now that approach is silly. It's much easier to work iteratively, change one thing, change another, until you have exactly what you need, in a much faster and more creative session.
So yeah you could use LLMs as a compiler, but it's so much more engaging not to.

by cschep

4 subcomments

Since LLM’s aren’t deterministic isn’t it impossible? What would keep it from iterating back and forth between two failing states forever? Is this the halting problem?

by fedeb95

1 subcomments

> Democratize access to engineering
```
    You don't need as specialized skillsets to build complex apps, you just need to know how to put context together and iterate
```
I feel it is exactly the opposite. AI helps specialists iterate faster, knowing what they are doing. Who doesn't know the details will stumble upon problems unsolvable by AI iteration. Who knows the details can step in where AIs fail.

by baalimago

9 subcomments

I've had the same exact thought! The reason why we've moved from higher to higher level of programming language is to make it easier for humans to describe to the machine what we want it to do. That's why languages are semantically easier and easier to read js > cpp > c > assembly > machine code (subjectively, yes yes, you get the point). It makes perfect sense to believe that natural language interpreted by an LLM is the next step in this evolution.
My prediction: in 10 years we'll see LLMs generate machine code directly, just like a normal compiler. The programming language will be the context provided by the context engineer.

by fxj

0 subcomment

That not only works for compilation but also for more general code transformations like code parallelization with OpenMP in C and Fortran, Array-of-lists to List-of-arrays, or transform python code to parallel C-code and make a python module out of it.
I have created some pipelines this way where the LLM generates input files for a molecular dynamics code and write a python script for execution on a HPC system.

by iamgvj

0 subcomment

There was a literal LLM compiler model that was released last year by Meta.
https://arxiv.org/abs/2407.02524

by UltraSane

0 subcomment

This really only works well if you have a TLA+ style formal model of the algorithm and can use it to generate lots of unit tests.

0 subcomment

by satisfice

0 subcomment

by pjmlp

0 subcomment

Not necessarly this, however I am quite convinced that AI based tooling will be the evolution of compilers.
The current way to generate source code of existing programming languages, is only a transition step, akin to how early compilers always generated Assembly that was further processed by an existing Assembler.
Eventually most developers don't even know the magic incantations to spew Assembly of their compilers, including JITs, it has become a dark art for compiler engineers, game developers and crypto folks.
People that used to joke about COBOL, would be surprised how much effort is being spent in prompt engineering.

by careful_ai

0 subcomment

Love this framing—treating LLMs like compilers captures how engineers mentally iterate: code, check, refine. It’s not about one-shot prompts. It’s a loop of design, compile, analyze, debug. That mindset shift—seeing LLMs as thought compilers—might be the missing link for real developer adoption.