Nanolang: A tiny experimental language designed to be targeted by coding LLMs
github.com182 points by Scramblejams 17 hours ago
182 points by Scramblejams 17 hours ago
At this point, I am starting to feel like we don’t need new languages, but new ways to create specifications.
I have a hypothesis that an LLM can act as a pseudocode to code translator, where the pseudocode can tolerate a mixture of code-like and natural language specification. The benefit being that it formalizes the human as the specifier (which must be done anyway) and the llm as the code writer. This also might enable lower resource “non-frontier” models to be more useful. Additionally, it allows tolerance to syntax mistakes or in the worst case, natural language if needed.
In other words, I think llms don’t need new languages, we do.
What we need is a programming language that defines the diff to be applied upon the existing codebase to the same degree of unambiguity as the codebase itself.
That is, in the same way that event sourcing materializes a state from a series of change events, this language needs to materialize a codebase from a series of "modification instructions". Different models may materialize a different codebase using the same series of instructions (like compilers), or say different "environmental factors" (e.g. the database or cloud provider that's available). It's as if the codebase itself is no longer the important artifact, the sequence of prompts is. You would also use this sequence of prompts to generate a testing suite completely independent of the codebase.
I am working on that https://github.com/gritzko/librdx Conflictless merge and overlay branches (ie freely attachable/detachable by a click). That was the pie-in-the-sky of the CRDT community for maybe 15 years. My current approach is RDX tree CRDT effectively mapping to the AST tree of the program. Like CRDT DOM for the AST, because line based diffs are too clumsy for that.
Back in the day, JetBrains tried revision-controlling AST trees or psi-nodes in their parlance. That project was cancelled, as it became a research challenge. That was 10 years ago or so. At this point, things may work out well, time will tell.
Was it cancelled? I thought MPS works that way.
I meant specifically revision control. JetBrains' school of thought is very much AST-centric, yes.
I think this could be very useful even for regular old programming. We could treat the diffs to the code as the main source of truth (instead of the textual snapshot each diff creates).
Jonathan Edwards (Subtext lang) has a lot of great research on this.
I think this confuses two different things:
- LLMs can act as pseudocode to code translators (they are excellent at this)
- LLMs still create bugs and make errors, and a reasonable hypothesis is at a rate in direct proportion to the "complexity" or "buggedness" of the underlying language.
In other words, give an AI a footgun and it will happily use it unawares. That doesn't mean however it can't rapidly turn your pseudocode into code.
None of this means that LLMs can magically correct your pseudocode at all times if your logic is vastly wrong for your goal, but I do believe they'll benefit immensely from new languages that reduce the kind of bugs they make.
This is the moment we can create these languages. Because LLMs can optimize for things that humans can't, so it seems possible to design new languages to reduce bugs in ways that work for LLMs, but are less effective for people (due to syntax, ergonomics, verbosity, anything else).
This is crucially important. Why? Because 99% of all code written in the next two decades will be written by AI. And we will also produce 100x more code than has ever been written before (because the cost of doing it, has dropped essentially to zero). This means that, short of some revolutions in language technology, the number of bugs and vulnerabilities we can expect will also 100x.
That's why ideas like this are needed.
I believe in this too and am working on something also targeting LLMs specifically, and have been working on it since Mid to Late November last year. A business model will make such a language sustainable.
Say you have this new language, with only a tiny amount of examples of there. How do the SOTA labs train on you're language? With sufficient examples, it can generate code which gets compiled and then run and that gets fed into a feedback loop to improve upon, but how do you get there? How do you bootstrap that? Nevermind the dollar cost, how does it offer something above having an LLM generate code in python or JavaScript, then having it rewrite it in golang/rust/c++ as needed/possible for performance or whatever reason?
It sounds like your plan is for it to write fewer bugs in NewLang, but, well, that seems a bit hard to achieve in the abstract. From bugs I've fixed in generated code, early LLM, it was just bad code. Multiple variables for the same thing, especially. Recently they've gotten better at that, but it still happens.
For a concrete example, any app dealing with points in time. Which sometimes have a date attached but sometimes do not. And also, what are timezones. The complexity is there because it depends on what you're trying to do. An alarm clock is different than a calendar is different than a pomodoro timer. How are you going to reduce the bugged-ed-ness of that without making one of those use cases more complicated than need be, given access to various primitives.
Your hypothetical misses praxis: in my experience LLM can pick up any new syntax with ease. From a few examples, it can generate more. With a compiler (even partial on limited syntax), it can correct. It soon becomes fluent simply from the context of your codebase. You don't need to "train" an LLM to recognize language syntax. It's effortless for it to pick it up.
Or, maybe my lanng just had LLM-easy syntax - which would be good - but I think this is more just par for the course for LLMs, bud.
I'm also looking at this topic right now.
I think you're right within limits but the issue is semantics and obscure features. If the language differs from existing languages in only trivial ways, then LLMs can pick it up quickly. But then the value of such a language is trivial. If you deviate in bigger ways, it's harder to properly use just based on pre-existing code.
Here's a simple case study: Kotlin is semantically Java with a more concise syntax, but part of what makes it more concise is the Kotlin standard library adds a lot of utility methods to Java. Many utility methods are only needed rarely. LLMs can write competent Kotlin because they read the user guide and saw millions of examples in their training set, but if they were trying to learn exclusively from small examples in their context window, they wouldn't know about those obscure utilities and would never use them. Much of the benefit would be lost.
Given this, I see a few ways forward:
1. Just give up on designing new programming languages. Languages are user interfaces but the user is now an LLM with near infinite patience, so who cares if they aren't ideal. If the LLM has to brute force a utility method every single time instead of using a standard library... ok. Whatever. This would parallel what happened with CPU ISAs. There are very few of them today, they don't matter much and they're designed in ways that only machines can handle all the details, because everyone codes to higher level languages and compilers write all the assembly.
2. Define new languages as a delta on top of some well known initial language, ensuring that the language definition always fits inside a prompt as a skill. In this world we don't bother with new syntaxes anymore unless that syntax change encodes significant new semantics, because it's not worth wasting tokens showing the LLM what to do. Everything is just an extension to Python, in this world. The line between new languages and new libraries becomes increasingly blurred as runtimes get more powerful and flexible.
3. New languages have to come with their own fine tuned and hosted coding LLM. Maybe that's even a way to monetize new language creation.
4. The big model firms offer a service where you can pay to get your data into the training set. Then you use the giant prompt+delta mechanism to get an LLM to generate a textbook of sample code, pay to get it into the training set, wait six months for another foundation model run and then your language becomes usable.
Of these I think (2) is currently the most practical.
This sounds academic, like a thought experiment. I have experience and can tell you this is not the case. I am using a significantly different language and the LLMs have 0 problem using it.
There's likely challenges here, but it's not the ones you're seeing so far.
OK, that's valuable to know, but how is your language different? You were discussing syntax previously. How well does the LLM handle your language's different standard library and how big is it?
OK, I'll bite :) Syntax is easy for LLMs, but the syntax is a bit different to normal langs. Underlying that the runtime is also unique in some ways, explicit abort (no limping failure states) on error, world modeling via union types, a unique auditable concurrency model, plus a "no stdlib / no C / no FFI" approach that means we are AOT to direct syscalls/kernel32 across linux/macos/win. Eventually likely add some FFI in a controlled way but as soon as you do you import all the vulns, so I'm doing TLS via execve curl first (I consider that safer/saner than just using openssl), but have a working net, and http, plus math, process, etc etc. It's small and focused, only around 45KSLOC with a compiler frontend, minimal IR and arch/os related backends emitting assembly (no LLVM). There's more but that's an overview.
Ah, people are starting to see the light.
This is something that could be distilled from some industries like aviation, where specification of software (requirements, architecture documents, etc.) is even more important that the software itself.
The problem is that natural language is in itself ambiguous, and people don't really grasp the importance of clear specification (how many times I have repeated to put units and tolerances to any limits they specify by requirements).
Another problem is: natural language doesn't have "defaults": if you don't specify something, is open to interpretation. And people _will_ interpret something instead of saying "yep I don't know this".
> The problem is that natural language is in itself ambiguous
This is literally what software developers are actually paid to do. They are not paid to write code. This is reinventing software development.
You can use LLMs as specification compilers. They are quite good at finding ambiguities in specs and writing out lists of questions for the author to answer, or inferring sensible defaults in explicitly called out ways.
we have some markup for architectures like - d2lang, sequencediagram.org's, bpmn.io xmls (which are OMG XMLs), so question is - can we master these, and not invent new stuf for a while?
p.s. a combination of the above fares very well during my agentic coding adventures.
I’ve been on a similar train of thought. Just last weekend I built a little experiment, using LLMs to highlight pseudocode syntax:
This is the approach that Agint takes. We inference the structure of the code first top down as a graph, then add in types, then interpret the types as in out function signatures and then "inpaint" the functions for codegen.
coding in latex and then translating to the target via llm works remarkably well nowadays
I'm actually building this, will release it early next month. I've added a URL to watch to my profile (should be up later this week). It will be Open Source.
>>new ways to create specifications.
Thats again programming languages. Real issue with LLMs now is it doesn't matter if it can generate code quickly. Some one still has to read, verify and test it.
Perhaps we need a need a terse programming language. Which can be read quickly and verified. You could call that specification.
This specification argument seems to boil down to: what if we used Haskell to describe systems to LLMs?
Many of our traditional functional languages, ML family in particular, let you write hyper concise expressions (pure math if you’re in to that sort of thing), craft DSLs of unlimited specifiable power (‘makeTpsReportWith “new cover page format”’), and also in natural language (function names like `emptied cart should have zero items`).
I think if we did that and leveraged the type systems of those languages and the systematic improvements we see from ADTs and pattern matching in those languages, combined with a specification first approach like TDD, that we’d have a great starting point to have an LLM generate the rest of the system perfectly.
… yes, that is just writing Haskell/OCaml/F# with extra steps.
… yes, that level of specification is also the point with those languages where your exploratory type-diddling suddenly goes ‘presto’ and you magically have a fully functioning system.
I guess I’m old-fashioned, but sometimes I wonder if compilers are good for what they’re good for.
Yes, essentially a higher level programming language than what we currently have. A programming language that doesn't have strict syntax, and can be expressed with words or code. And like any other programming language, it includes specifications for the tests and expectations of the result.
The programming language can look more like code in parts where the specification needs to be very detailed. I think people can get intuition about where the LLM is unlikely to be successful. It can have low detail for boilerplate or code that is simple to describe.
You should be able to alter and recompile the specification, unlike the wandering prompt which makes changes faster than normal version control practices keep up with.
Perhaps there's a world where reading the specification rather than the compiled code is sufficient in order to keep cognitive load at reasonable levels.
At very least, you can read compiled code until you can establish your own validation set and create statistical expectations about your domain. Principally, these models will always be statistical in nature. So we probably need to start operating more inside that kind of framework if we really want to be professional about it.
We already have exceptionally high level languages, like Inform7 [0]. The concept doesn't work all that well. Terseness is a value. Its why we end up with so many symbol-heavy languages. Yes, there are tradeoffs, but that is the whole of computer science.
We didn't end up with Lean and Rust, for a lack of understanding in how to create strong specifications. Pascal-like languages fell out of favour, despite having higher readability.
Simply put whatever you write should produce the same output regardless of how many times you execute it. The more verbose you make it, the more pointless it becomes.
More terse the better.
So in this case an LLM would just be a less-reliable compiler? What's the point? If you have to formally specify your program, we already have tools for that, no boiling-the-oceans required
llm works great in closed loop so they can self correct but we don't have a reliable way to lint and test specs we need a new language for that
Go read ai2027 and then be ashamed of yourself /s
But seriously, llms can transmit ideas to each other through English that we do understand, we are screwed if it’s another language lol
Developed by Jordan Hubbard of NVIDIA (and FreeBSD).
My understanding/experience is that LLM performance in a language scales with how well the language is represented in the training data.
From that assumption, we might expect LLMs to actually do better with an existing language for which more training code is available, even if that language is more complex and seems like it should be “harder” to understand.
I don’t think that assumption holds. For example, only recently have agents started getting Rust code right on the first try, but that hasn’t mattered in the past because the rust compiler and linters give such good feedback that it immediately fixes whatever goof it made.
This does fill up context a little faster, (1) not as much as debugging the problem would have in a dynamic language, and (2) better agentic frameworks are coming that “rewrite” context history for dynamic on the fly context compression.
> that hasn’t mattered in the past because the rust compiler and linters give such good feedback that it immediately fixes whatever goof it made.
This isn't even true today. Source: heavy user of claude code and gemini with rust for almost 2 years now.
Yeah, I have zero problem getting Opus 4.5 to write high-quality Rust code. And I'm picky.
I have no problems with rust and Claude Code, and I use it on a daily basis.
Blackpill is that, for this reason, the mainstream languages we have today will be the final (human-designed) languages to be relevant on a global scale.
Eventually AIs will create their own languages. And humans will, of course, continue designing hobbyist languages for fun. But in terms of influence, there will not be another human language that takes the programming world by storm. There simply is not enough time left.
A lot of this depends on your workflow. A language with great typing, type checking and good compiler errors will work better in a loop than one with a large surface overhead and syntax complexity, even if it's well represented. This is the instinct behind, e.g. https://github.com/toon-format/toon, a json alternative format. They test LLM accuracy with the format against JSON, (and are generally slightly ahead of JSON).
Additionally just the ability to put an entire language into context for an LLM - a single document explaining everything - is also likely to close the gap.
I was skimming some nano files and while I can't say I loved how it looked, it did look extremely clear. Likely a benefit.
Thanks for sharing this! A question I've grappled with is "how do you make the DOM of a rendered webpage optimal for complex retrieval in both accuracy and tokens?" This could be a really useful transformation to throw in the mix!
I mostly agree, and I think a combination of good representation and tooling that lets it self-correct quickly will do better than new language in the short term.
In the long term I expect it won't matter - already GPT3.5 was able to reason about the basic semantics of programs in languages "synthesised" zero-shot in context by just describing it as a combination of existing languages (e.g. "Ruby with INTERCAL's COME FROM") or by providing a grammar (e.g. simple EBNF plus some notes on new/different constructs) reasonably well and could explain what a program written in a franken-language it had not seen before was likely to do.
I think long before there is enough training data for a new language to be on equal grounds in that respect, we should expect the models to be good enough at this that you could just provide a terse language spec.
But at the same time, I'd expect the same improvement to future models to be good enough at working with existing languages that it's pointless to tailor languages to LLMs.
It's not just how well the language is represented. Obscure-ish APIs can trip up LLMs. I've been using Antigravity for a Flutter project that uses ATProto. Gemini is very strong at Dart coding, which makes picking up my 17th managed language a breeze. It's also very good at Flutter UI elements. It was noticeably less good at ATProto and its Dart API.
The characteristics of failures have been interesting: As I anticipated it might be, an over ambitious refactoring was a train wreck, easily reverted. But something as simple as regenerating Android launcher icons in a Flutter project was a total blind spot. I had to Google that like some kind of naked savage running through the jungle.
I have a vibe coded fantasy console. Getting Doom running on it was easy.
Getting the Doom sound working on it involved me setting there typing "No I can't hear anything" over and over until it magically worked...
Maybe I should have written a helper program to listen using the microphone or something.
> My understanding/experience is that LLM performance in a language scales with how well the language is represented in the training data.
This isn't really true. LLMs understand grammars really really well. If you have a grammar for your language the LLM can one-shot perfect code.
What they don't know is the tooling around the language. But again, this is pretty easily fixed - they are good at exploring cli tools.
I think it's depressingly true of any novel language/framework at this point, especially if they have novel ideas.
I wonder if there is a way to create a sort of 'transpilation' layer to a new language like this for existing languages, so that it would be able to use all of the available training from other languages. Something that's like AST to AST. Though I wonder if it would only work in the initial training or fine-tuning stage.
Not my experience, honestly. With a good code base for it to explore and good tooling, and a really good prompt I've had excellent results with frankly quite obscure things, including homegrown languages.
As others said, the key is feedback and prompting. In a model with long context, it'll figure it out.
But isn't this inefficient since the agent has to "bootstrap" its knowledge of the new language every time it's context window is reset?
No, it gets it “for free” just by looking around when it is figuring out how to solve whatever problem it is working on.
Yeah, I've had Claude work on my buggy, incomplete Ruby compiler written (mostly) in Ruby, which uses an s-expression like syntax with a custom "mini language" to implement low-level features that can't be done (or is impractical to do) in pure Ruby, and it only had minor problems with the s-expression language that was mostly fixed with a handful of lines in CLAUDE.md (and were, frankly, mostly my fault for making the language itself somewhat inconsistent) and e.g. when it write a bigint implementation, I had to "tell it off" for too readily resorting to the s-expression syntax since it seemed to "prefer it" over writing high-level code in Ruby.
Even 3 years ago, GH Copilot, hardly the most intelligent of LLMs was suggesting/writing bytecode in my custom VM, writing full programs in bytecode for a custom VM just by looking at a couple examples.
That's when I smelled that things were getting a little crazy.
easy enough to solve with RL probably
There is no RL for programming languages. Especially ones w/ no significant amount of code.
I guess the op was implying that is something fixable fairly easily?
(Which is true - it's easy to prompt your LLM with the language grammar, have it generate code and then RL on that)
Easy in the sense of "it is only having enough GPUs to RL a coding capable LLM" anyway.
If you can generate code from the grammar then what exactly are you RLing? The point was to generate code in the first place so what does backpropagation get you here?
Go read the DeepSeek R1 paper
Why would I do that? If you know something then quote the relevant passage & equation that says you can train code generators w/ RL on a novel language w/ little to no code to train on. More generally, don't ask random people on the internet to do work for you for free.
Your other comment sounded like you were interested in learning about how AI labs are applying RL to improve programming capability. If so, the DeepSeek R1 paper is a good introduction to the topic (maybe a bit out of date at this point, but very approachable). RL training works fine for low resource languages as long as you have tooling to verify outputs and enough compute to throw at the problem.
I went looking for a single Markdown file I could dump into an LLM to "teach" it the language and found this one:
https://github.com/jordanhubbard/nanolang/blob/main/MEMORY.m...
Optimistically I dumped the whole thing into Claude Opus 4.5 as a system prompt to see if it could generate a one-shot program from it:
llm -m claude-opus-4.5 \
-s https://raw.githubusercontent.com/jordanhubbard/nanolang/refs/heads/main/MEMORY.md \
'Build me a mandelbrot fractal CLI tool in this language'
> /tmp/fractal.nano
So I fired up Claude Code inside a checkout of the nanolang and told it how to run the compiler and let it fix the problems... which DID work. Here's that transcript:
https://gisthost.github.io/?9696da6882cb6596be6a9d5196e8a7a5...
And the finished code, with its output in a comment: https://gist.github.com/simonw/e7f3577adcfd392ab7fa23b1295d0...
So yeah, a good LLM can definitely figure out how to use this thing given access to the existing documentation and the ability to run that compiler.
Oh, wow. I thought the control flow from the readme was a little annoying with the prefix -notation for bigger/smaller than;
# Control flow
if (> x 0) {
(println "positive")
} else {
(println "negative or zero")
}
# Gradient: " .:-=+*#%@"
let gradient: string = " .:-=+*#%@"
let gradient_len: int = 10
let idx: int = (/ (* iter gradient_len) max_iter)
if (>= idx gradient_len) {
return "@"
} else {
if (== idx 0) {
return " "
} else {
if (== idx 1) {
return "."
} else {
if (== idx 2) {
return ":"
} else {
if (== idx 3) {
return "-"
} else {
if (== idx 4) {
return "="
} else {
if (== idx 5) {
return "+"
} else {
if (== idx 6) {
return "*"
} else {
if (== idx 7) {
return "#"
} else {
if (== idx 8) {
return "%"
} else {
return "@"
}
}I mean for all intents and purposes this language is designed for use by LLM's, not humans, and the AI probably won't complain that a switch-case statement is missing. ;)
> scream for a case/switch-statement
Maybe I’m missing some context, but all that actually should be needed in the top-level else block is ‘gradient[idx]’. Pretty much anything else is going to be longer, harder to read, and less efficient.
True, with early return - there's no need to actually nest with else.
Logically this still would be a case/switch though...
The point was that logically it would be an array lookup by index.
There's no need for any conditional construct here whatsoever.
You'll note it has already constructed a string in the right order to do that, but then copped out with the if-else.
True enough. On that note, I had a look at the language reference - there's arrays - but also this:
(char_at s index) # Get ASCII value at index (0-based)
(string_from_char code) # Create string from ASCII value
If you are planning to write so many if else statements. You might as well write Prolog.
I think you need to either feed it all of ./docs or give your agent access to those files so it can read them as reference. The MEMORY.md file you posted mentions ./docs/CANONICAL_STYLE.md and ./docs/LLM_CORE_SUBSET.md and they in turn mention indirectly other features and files inside the docs folder.
Yeah, I think you're right about that.
The thing that really unlocked it was Claude being able to run a file listing against nanolang/examples and then start picking through the examples that were most relevant to figuring out the syntax: https://gisthost.github.io/?9696da6882cb6596be6a9d5196e8a7a5...
But are you losing horsepower of the LLM available to problem solving on a given task by doing so?
Maybe a little, but Claude has 200,000 tokens these days and GPT-5.2 has 400,000 - there's a lot of space.
True. You would know this better but are you also burning "attention" by giving it a new language? Rather than use its familiar Python pathways it needs to attend more to generate the unseen language. It needs to KV across from the language spec to the language to the goal. Rather than just speak the Python or JS it is uses to speaking.
The required-test-per-function is sort of interesting. But it's not enforced that the test does anything useful, is it?
So I wonder how exhausting would it be to write in a language that required, for all functions, that they are tested with 100% path coverage.
Of course, this by itself wouldn't still be equivalent to proving the code, but it would probably point people to the corner cases of code quite rapidly. Additionally it would make it impossible to have code that cannot be tested with 100% path coverage due to static relationships within it, that are not (or cannot be) expressed in the type system, e.g. if (foo) { if (!foo) {..} }.
And would such a language need to have some kind of dynamic dependency injection mechanism for mocking the tests?
Author has a Wikipedia page about him linked from his GitHub profile: https://en.wikipedia.org/wiki/Jordan_Hubbard
Summary:
- Co-created FreeBSD.
- Led UNIX technologies at Apple for 13 years
- iXSystems, lead FreeNAS
- idk something about Uber
- Senior Director for GPU Compute Software at NVIDIA
For whatever it’s worth.
a.k.a. jkh. That's a blast from the past. Back in the early FreeBSD days, Jordan was fielding mailing list traffic and holding the project together as people peppered the lists with questions, trying to get their systems running with their sundry bits of hardware. I wondered when he slept.
Apparently he did as well[1]: "The start of the 2.0 ports collection. No sup repository yet, but I'll make one when I wake up again.. :)" Submitted by: jkh Aug 21, 1994
[1] https://github.com/freebsd/freebsd-ports/commit/7ca702f09f29...
Interesting commit starting Ports 2.0. Three version of bash, four versions of Emacs, plus jove.
There’s both efficacy and token efficiency to consider here.
Seems unlikely for an out-of-distribution language to be as effective as one that’s got all the training data in the world.
Really needs an agent-oriented “getting started” guide to put in the context, and evals vs. the same task done with Python, Rust etc.
> Really needs an agent-oriented “getting started” guide to put in the context, and evals vs. the same task done with Python, Rust etc.
It has several such documents, including a ~1400 line MEMORY.md file referencing several other such files, a language specification, a collection of ~100 documents containing just about every thought Jordan has ever had about the entire language and the evolution of its implementation, and a collection of examples that includes an SDL2 based OpenGL program.
Obviously, jkh clearly understands the need to bootstrap LLMs on his ~5 month old, self-hosted solo programming language.
One novel part here is every function is required to have tests that run at compile time.
I'm still skeptical of the value add having to teaching a custom language to an LLM instead of using something like lua or python and applying constraints like test requirements onto that.
I'm not sure that it's novel but I'm skeptical about the noise to signal ratio for anything that is not an example.
I think that a real world file of source code will be either completely polluted by tests (they are way longer than the actual code they test) or become
fn process_order {
...
}
shadow process_order {
assert test_process_order
}
Pyret, a teaching language for CS, in the vein of Racket, does require testing by writing functions.
So, then if I want to use a certain terminal text editor to create a clone of it in nanolang, I'd end up typing nano nano.nano on the command line.
I might accidentally summon a certain person from Ork.
I can't seem to find in the repo what optimizes this language for LLMs and benchmarks for why it works?
The author's answers are toward the bottom of the README, https://github.com/jordanhubbard/nanolang?tab=readme-ov-file...
I think this kind of misses what's actually challenging with LLM code -- auditing it for correctness. LLMs are ~fine at spitting out valid syntax. Humans need to be able to read the output, though.
A language targeting an LLM might be well served with a lot of keywords, similar to a CISC instruction set, where keywords do specific things well. Giving it building blocks and having them piece together is likely to pay off.
It seems that something that does away with human friendly syntax and leans more towards a pure AST representation would be even better? Basically a Lisp but with very strict typing might do the trick. And most LLMs are probably trained on lots of Lisps already.
Generally seems a bad idea to have your LLM write languages you do not understand or write yourself
Doesn’t that apply to the OP as well?
Yes, I'm not going to fill my precious context with documentation for a programming language
This seems like a research dead end to me, the fundamentals are not there
It seems kind of silly that you can’t teach an LLM new tricks though, doesn’t it? This doesn’t sound like an intrinsic limitation and more an artifact of how we produce model weights today.
getting tricks embedded into the weights is expensive, it doesn't happen in a single pass
they's why we teach them new tricks on the fly (in-context learning) with instruction files
Seems like a simplified Rust with partial prefix notation (which the rationale that is better for LLMs is based on vibes really) that compiles to C. Similar language posted here not too long ago: Zen-C => more features, no prefix notation / Rue => no prefix notation, compiles directly to native code (no C target). Surprisingly compared to other LLM "optimized" languages, it isn't so much concerned about token efficiency.
I find Polish or Reverse Polish notation jarring after a lifetime of thinking in terms of operator precedence. Given that it's fairly rare to see, I wonder what about it would be more LLM-friendly. It does lend itself better to "tokenization" of a sort - if you want to construct operations from lots of smaller operations, for example if you're mutating genetic algorithms (a la Eureqa). But I've written code in the past to explicitly convert those kinds of operations back to infix for easier readability. I wonder if the LLMs in this case are expected to behave a bit like genetic algorithms as they construct things.
>It does lend itself better to "tokenization" of a sort - if you want to construct operations from lots of smaller operations [...]
That's an educated assumption to make. But therein lies the issue with every LLM "optimized" language, including those recent ones posted here oriented toward minimizing tokens. Assumptions, that are unvalidatable and unfalsifiable, about the kind of output LLMs synthesize/emit when that output is code (or any output to be real).
Just scanning through this, looks interesting and is totally needed, but I think it is missing showing future use-cases and discussions of decoding. So, for instance, it is all well and good to define a simple language focused on testing and the like, but what about live LLM control and interaction via a programming language? Sort of a conversation in code? Data streams in and function calls stream out with syntax designed to minimize mistakes in calls and optimize the stream? What I mean by this is special block declarations like:
``` #this is where functions are defined and should compile and give syntax errors ```
:->r = some(param)/connected(param, param, @r)/calls(param)<-:
(yeah, ugly but the idea is there) The point being that the behavior could change. In the streaming world it may, for instance, have guarantees of what executes and what doesn't in case of errors. Maybe transactional guarantees in the stream blocks compared to pure compile optimization in the other blocks? The point here isn't that this is the golden idea, but that we probably should think about the use cases more. High on my list of use cases to consider (I think)
- language independence: LLMs are multilingual and this should be multilingual from the start.
- support streaming vs definition of code.
- Streaming should consider parallelism/async in the calls.
- the language should consider cached token states to call back to. (define the 'now' for optimal result management, basically, the language can tap into LLM properties that matter)
Hmm... That is the top of my head thoughts at least.
Why "shadow" instead of "test"?
Thanks for all of the comments!
Quick reaction:
1. Nanolang is a total thought experiment. The key word its description is "experimental" - whether it's a Good experiment or a Bad experiment can be argued either way, especially by language purists!
2. Yes, it's a total Decorator Crab of a language. An unholy creation by Dr Frankenstein, yes! Those criticisms are entirely merited. It wasn't designed, it accreted features and was a fever dream I couldn't seem to stop having. I should probably take my own temperature.
3. I like prefix notation because my first calculator was an HP calculator (the HP 41C remains, to this day, my favorite calculator of ALL TIME). I won't apologize for that, but I DO get that it's not everybody's cup of tea! I do, however, use both vi and emacs now.
Umm. I think that about covers it. All of this LLM stuff is still incredibly young to me and I'm just firing a shotgun into the dark and listening to hear if I hit anything. It's going to be that way for a while for all of us until we figure out what works and what does not!
- jkh
Looks a bit like Rust. My peeve with Rust is that it makes error handling too much donkey work. In a large class of programs you just care that something failed and you want a good description of that thing:
context("Loading configuration from {file}")
Failed: Loading configuration from .config because: couldn't open file .config because: file .config does not exist.
It shouldn't be harder than a context command in functions. But somehow Rust conspires to require all this error type conversion and question marks. It it is all just a big uncomfortable donkey game, especially when you have nested closures forced to return errors of a specific type.
You just described how the popular "anyhow" and "snafu" crates implement error handling
I like your "context" proposal, because it adds information about developer intention to error diagnostics, whereas showing e.g. a call stack would just provide information about the "what?", not the "why?" to the end user facing an error at runtime.
(You should try to get something like that into various language specs; I'd love you to success with it.)
EDIT: typo fixed.
I'm a bit saddened that a language with 29 keywords is now considered tiny...
Really clean language where the design decisions have led to fewer traps (cond is a good choice).
It’s peculiar to see s-expressions mixed together with imperative style. I’ve been experimenting along similar lines - mixing s-expressions with ML style in the same dialect (for a project).
Having an agentic partner toiling away with the lexer/parser/implementation details is truly liberating. It frees the human to explore crazy ideas that would not have been feasible for a side/toy/hobby project earlier.
I had the same reaction to seeing S-expressions, but my conclusion was that this should just be a lisp. Why stop just at operators?
as i already wrote in an other comment. Where are the millions lines of code needed to train LLM in this Nanolang? LLM are like parrots. if you dont give them data to extract the statistic probability of the next word, you will not get any usefull output. LLM do not think, they can't learn without training data
That's an incorrect assumption. You can get quite far with some skill documents and some examples in combination with tools to compile and run your code. The LLM will train itself on the fly based on the feedback from these tools.
I feel this could be achieved better with Golang or Kotlin and a custom linter that enforces parentheses around each expression term to make precedence explicit, and enforce each function has at least one test. Although I guess neither of those languages has free interop with C, they are close. And Go doesn’t have unions :’(
An LLM targeting language and no token efficiency?
I'd rather see a programing language optimized for "few tokens". Something like toon, but for code.
I feel like the time for this was two years ago, and LLMs are now less bothered by remembering syntax than I am. It's a nice lisp-y syntax though.
opus is currently the only one that can code rust, but if you give it symbol resolution there is quite literally nothing better. The type system in rust is incredibly powerful and llms are great (just opus for now) at utilizing it.
Codex generates solid Rust in my experience. Just needs a little style guidance
Why NanoLang? NanoLang solves three problems:
LLM Code Generation - Unambiguous syntax reduces AI errors
Testing Discipline - Mandatory tests improve code quality
Simple & Fast - Minimal syntax, native performance
Design Philosophy:
Minimal syntax (18 keywords vs 32 in C)
One obvious way to do things
Tests are part of the language, not an afterthought
Transpile to C for maximum compatibility
ehh. i dont think the overhead of inventing a new language makes up for the lack of data around it. in fact if you're close enough to rust/c then llms are MORE likely to make up stuff from their training data and screw up your minimal language.
(pls argue against this, i want to be proven wrong)
there was nothing nano about that language at all if you want true nano drop return keywoards or just go APL
Almost no positive comments. I for one find this a great try, a very interesting project. I hope this kind of experiments gain track.
> tiny … language designed to be targeted by coding LLMs
so like Go?
> Key Features; Prefix Notation
wow
NEXT!
Every new language pet project these days claims to be "designed for LLM's", lol. Don't read too much into it. The only language that's really designed for LLM is COBOL, because it was written to read just like English natural language and LLM's are trained by reading lots of English language books.
Isn't this essentially just WebAssembly or Lisp?
It looks like a Frankenstein's abomination that has c-like function signatures and structs with Sexpr function bodies and this will anger some homomorphism nerds. I love it.
I mean WAT (WebAssembly Text Format) is essentially like that
You mean WAT has C-like struct and function signature syntax? I'm not seeing it.
Where will the training data come from?
Why did you pick shadow as a keyword for a testing block?
Looks nice!
This is kinda frustrating, docs are LLM generated, no explanation or justification for how this was designed to be targeted for LLMs.