LLMs could be, but shouldn't be compilers
alperenkeles.com
alperenkeles.com
It all feels to me like the guys who make videos of using using electric drills to hammer in a nail - Sure, you can do that, but it is the wrong tool for the job. Everyone knows the phrase: "When all you have is a hammer, everything looks like a nail." But we need to also keep in mind the other side of that coin: "When all you have is nails, all you need is a hammer." LLMs are not a replacement for everything that happens to be digital.
Often it seems like tech maximalists are the most against tech reliability.
I suggest when their pointer dereferences, it can go a bit forward or backwards in memory as long as it is mostly correct.
Imagine that - you got your project done ahead of schedule (which looks great on your OKRs) AND finally achieved your dream of no longer being dependent on those stupid overpaid, antisocial software engineers, and all it cost you was the company's reputation. Boeing management would be proud.
Lots of business leaders will do the math and decide this is the way to operate from now on.
No. LLMs are undefined behavior.
But most LLM services on purpose introduce randomness, so you don’t get the same result for the same input you control as a user.
Then my job became I am assigned a larger implementation and depending on how large the implementation was, I had to design specifications for others to do some or all of the work and validate the final product for correctness. I definitely didn’t pore over every line of code - especially not for front end work that I stopped doing around the same time.
The same is true for LLMs. I treat them like junior developers and slowly starting to treat them like halfway competent mid level ticket takers.
Humans, in all their non deterministic brain glory, long ago realized they don't want their software to behave like their coworkers after a couple of margaritas.
"Deterministic" is not the the right constraint to introduce here. Plenty of software is non-deterministic (such as LLMs! But also, consensus protocols, request routing architecture, GPU kernels, etc) so why not compilers?
What a compiler needs is not determinism, but semantic closure. A system is semantically closed if the meanings of its outputs are fully defined within the system, correctness can be evaluated internally and errors are decidable. LLMs are semantically open. A semantically closed compiler will never output nonsense, even if its output is nondeterministic. But two runs of a (semantically closed) nondeterministic compiler may produce two correct programs, one being faster on one CPU and the other faster on another. Or such a compiler can be useful for enhancing security, e.g. programs behave identically, resist fingerprinting.
Nondeterminism simply means the compiler selects any element of an equivalence class. Semantic closure ensures the equivalence class is well‑defined.
I am not. To me that describes a debugging fiasco. I don't want "semantic closure," I want correctness and exact repeatability.
On a practical level, existing implementations are nondeterministic because they don't take care to always perform mathematically commutative operations in the same order every time. Floating-point arithmetic is not commutative, so those variations change the output. It's absolutely possible to fix this and perform the operations in the same order every time, implementors just don't bother. It's not very useful, especially when almost everything runs with a non-zero temperature.
I think the whole nondeterminism thing is overblown anyway. Mathematical nondeterminism and practical nondeterminism aren't the same thing. With a compiler, it's not just that identical input produces identical output. It's also that semantically identical input produces semantically identical output. If I add an extra space somewhere whitespace isn't significant in the language I'm using, this should not change the output (aside from debug info that includes column numbers, anyway). My deterministic JSON decoder should not only decode the same values for two runs on identical JSON, a change in one value in the input should produce the same values in the output except for the one that changed.
LLMs inherently fail at this regardless of temperature or determinism.
Meanwhile, you press the "shuffle" button, and code-gen creates different code. But this isn't necessarily the part that's supposed to be reproducible, and isn't how you actually go about comparing the output. Instead, maybe two different rounds of code-generation are "equal" if the test-suite passes for both. Not precisely the equivalence-class stuff parent is talking about, but it's simple way of thinking about it that might be helpful
No, a compiler needs determinism. The article is quite correct on this point: if you can't trust that the output of a tool will be consistent, you can't use it as a building block. A stochastic compiler is simply not fit for purpose.
What matters is observable execution.
That a compiler might pick among different specific implementations in the same equivalency class is exactly what you want a multi-architecture optimizing compiler to do. You don't want it choosing randomly between different optimization choices within an optimization level, that would be non-deterministic at compile time and largely useless assuming that there is at most one most optimized equivalent. I always want the compiler to choose to xor a register with itself to clear it if that's faster than explicitly setting it to zero if that makes the most sense to do given the inputs/constraints.
https://thinkingmachines.ai/blog/defeating-nondeterminism-in...
if you rewrote the math to be all fixed point precision on big ints, i think you would still get the useful LLM results?
if somebody really wanted to make a compiler in an LLM, i dont think that nondetermism is problem
id really imagine an llm compiler being a set of specs, dependency versions, and test definitions to use though, and you'd introduce essential nondetermism by changing a version number, even if the only change was the version name from "experimental" to "lts"
There are legitimate compiler use cases e.g. search‑based optimization, superoptimization, diversification etc where reproducibility is not the main constraint. It's worth leaving conceptual space for those use cases rather than treating deterministic output as a defining property of all compilers
You are attempting to hedge and leave room for a non-deterministic compiler, presumably to argue that something like vibe-compilation is valuable. However, you've offered no real use cases for a non-deterministic compiler, and I assert that such a tool would largely be useless in the real world. There is already a huge gap between requirements gathering, the expression of those requirements, and their conversion into software. Adding even more randomness at the layer of translating high level programming languages into low level machine code would be a gross regression.
or does your binary always come out differently each time you compile the same file??
You can try it. try to compile the same file 10 times and diff the resultant binaries.
Now try to prompt a bunch of LLMs 10 times and diff the returned rubbish.
There's this really good blog post about how autovectorization is not a programming model https://pharr.org/matt/blog/2018/04/18/ispc-origins
The point is that you want to reliably express semantics in the top level language, tool, API etc. because that's the only way you can build a stable mental model on top of that. Needing to worry about if something actually did something under the hood is awful.
Now of course, that depends on the level of granularity YOU want. When writing plain code, even if it's expressively rich in the logic and semantics (e.g. c++ template metaprogramming), sometimes I don't necessarily care about the specific linker and assembly details (but sometimes I do!)
The issue I think is that building a reliable mental model of an LLM is hard. Note that "reliable" is the key word - consistent. Be it consistently good or bad. The frustrating thing is that it can sometimes deliver great value and sometimes brick horribly and we don't have a good idea for the mental model yet.
To constrain said possibility space, we tether to absolute memes (LLMs are fully stupid or LLMs are a superset of humans).
Idk where I'm going with this
There’s even efforts to guarantee this for many packages on Linux - it’s a core property of security because it lets you validate that the compilation process or environment wasn’t tampered with illicitly by being able to verify by building from scratch.
Now actually managing to fix all inputs and getting deterministic output can be challenging, but that’s less to do with the compiler and more to do with the challenge of completely taking the entire environment (the profile you are using for PGO, isolating paths on the build machine being injected into the binary, programs that have things in their source or build system that’s non deterministic (e.g. incorporating the build time into the binary)
> PGO seems like it ought to have a random element.
PGO should be deterministic based on the runs used to generate the profile. The runs are tracking information that should be deterministic--how many times does the the branch get taken versus not taken, etc. HWPGO, which relies on hardware counters to generate profiling information, may be less deterministic because the hardware counters end up having some statistical slip to them.
Reproducible builds are oft considered a desirable goal, but not having them is hardly a bug. Not even some of the most well known compilers out there, like gcc and clang, are deterministic by default — and not by accident; in those cases you can enable determinism if you wish.
Hence why it is hard to do benchmarks with various kinds of GC and dynamic compilers.
You can't even expect deterministic code generation for the same source code across various compilers.
They are designed to be where temperature=0. Some hardware configurations are known defy that assumption, but when running on perfect hardware they most definitely are.
What you call compilers are also nondeterministic on 'faulty' hardware, so...
To say the least, this is garbage compared to compilers
When isn't that true?
int main() {
printf("Continue?\n");
}
int main() {
printf("Continue?\n");
printf("Continue?\n");
}
With LLMs the output depends on the phases of the moon.
As with LLMs, unless you ask for the output to be nondeterministic. But any compiler can be made nondeterministic if you ask for it. That's not something unique to LLMs.
> With LLMs the output depends on the phases of the moon.
If you are relying on a third-party service to run the LLM, quite possibly. Without control over the hardware, configuration, etc. then there is all kinds of fuckery that they can introduce. A third-party can make any compiler nondeterministic.
But that's not a limitation of LLMs. By design, they are deterministic.
Not unique as in: no one makes their compilers deterministic, and you have to work to make a non-deterministic one. LLMs are non-deterministic by default, and you have to contort them to the point of uselessness to make them deterministic
> If you are relying on a third-party service to run the LLM, quite possibly. Without control over the hardware, configuration, etc.
Again. Even if you control everything, the only time they produce deterministic output is when they are completely neutered:
- workaround for GPUs with num_thread 1
- temperature set to 0
- top_k to 0
- top_p to 0
- context window to 0 (or always do a single run from a new session)
Go (gc) was designed for reproducible builds by default, so clearly that's not true, but you are right that it isn't the norm.
Even the most widely recognized and used compilers, like gcc, clang, even rustc, are non-deterministic by default. Only if you work hard and control all the variables (e.g. -frandom-seed) can you make these compilers deterministic.
It's fascinating that anyone on HN thinks that compilers converge on always being deterministic or always being non-deterministic. I thought we were supposed to know things about computers around here?
I was typing too fast. "No one makes their compilers nondeterministic.
> Even the most widely recognized and used compilers, like gcc, clang, even rustc, are non-deterministic by default.
Wat?
In which world are these compilers producing non-deterministic output if you run them again and again?
> It's fascinating that anyone on HN thinks that compilers converge on always being deterministic
It's called reality that is also very trivially verified.
> I thought we were supposed to know things about computers around here?
That's what I thought, too.
The one where deterministic iteration order (think use of unordered maps/sets, parallel execution, etc.) isn't considered a priority or is even seen as something to avoid as determinism here can lead to slower compiler times. Where there is use of heuristics that can have a "tie score" without effort to choose the winner deterministically. Where there is a need to support features like __DATE__ and __TIME__. So on and so on. It is a little out of the way so you've probably never heard of it, but its inhabitants call it "Earth".
Strange how literally no one is talking about this fictional non-determinism where the output of the compiler is completely different from run to run, but for the "LLMs are strictly deterministic" non-determinism is the norm and everyone expects it.
Edit also note how you went from "any compiler can be made nondeterministic if you ask for it." (compilers are deterministic, and you have to work to make them non-deterministic) to "most widely recognized and used compilers are non-deterministic by default."
There are specific things (undefined behavior) and specific things (like floating point precision between implementations) that may be specified as non-deterministic. To pretend that this is somehow equal to LLMs is not even being unrealistic. It's to exist on a plane orthogonal to our reality.
Oh, you are talking about program determinism. I don't know about fictional compilers, but gcc and clang will produce programs that behave differently from compile to compile where __DATE__ or __TIME__ is used. There are languages like Church that are explicitly designed for non-deterministic programs, so a Church compiler would obviously need to adhere to that. And, of course, ick has the mysterious -mystery flag!
But we were talking about compiler determinism. Where for a stable input the compiler produces a stable output. A non-deterministic compiler does not necessarily equate to a non-deterministic program. Obvious to anyone who has used a computer before, the structure of the binary produced by a compiler and the execution of that binary are separate concerns.
If you wanted to talk about something completely different why not start a new thread?
That's not non-determinism, is it?
> Where for a stable input the compiler produces a stable output. A non-deterministic compiler does not necessarily equate to a non-deterministic program
In case of LLMs that you're desperately trying to equate compilers to that's exactly what it is.
If I take a compiler and run it on an input in debug and release mode, with an extremely high degree of confidence I'll get the same working program (with the deterministic difference in debug logging etc.)
For most compilers determinism is such that people can often predict the final output (down to assembly instructions emitted by C++ compilers). Or target specific intermediate modes directly when needed.
If I take LLMs and change literally any variable from 0/0/0, I will get entirely unpredictable uncontrollable random output.
> If you wanted to talk about something completely different why not start a new thread?
Your vague and non-vague ad hominims are getting tiresome.
Adieu.
In the context of discussion, yes, as the only input is considered to be the source code. It is both non-deterministic in compiler output and in program function (two different builds of the same code will produce programs that do different things). Side channel inputs like time are not considered here. If you are trying to start a new thread again and you want to include time as an input in that discussion, then no. However, the same is true of LLMs. They are 100% always deterministic when all inputs are accounted for.
Obviously. Working computers fundamentally cannot operate non-deterministically, although in the real world can appear to when some or all inputs are not accounted for. The discussion is premised on non-determinism meaning "some inputs are ignored".
We have mechanisms for ensuring output from humans, and those are nothing like ensuring the output from a compiler. We have checks on people, we have whole industries of people whose whole careers are managing people, to manage other people, to manage other people.
with regards to predictability LLMs essentially behave like people in this manner. The same kind of checks that we use for people are needed for them, not the same kind of checks we use for software.
Those checks works for people because humans and most living beings respond well to rewards/punishment mechanisms. It’s the whole basis of society.
> not the same kind of checks we use for software.
We do have systems that are non deterministic (computer vision, various forecasting models…). We judge those by their accuracy and the likely of having false positive or false negatives (when it’s a classifier). Why not use those metrics?
LLM code completion compares unfavourably to the (heuristic, nigh-instant) picklist implementations we used to use, both at the low-level (how often does it autocomplete the right thing?) and at the high-level (despite many believing they're more effective, the average programmer is less effective when using AI tools). We need reasons to believe that LLMs are great and do all things, therefore we look for measurements that paint it in a good light (e.g. lines of code written, time to first working prototype, inclination to output Doom source code verbatim).
The reason we're all using (or pretending to use) LLMs now is not because they're good. It's almost entirely unrelated.
The whole benefit of computers is that they don't make stupid mistakes like humans do. If you give a computer the ability to make random mistakes all you have done is made the computer shitty. We don't need checks, we need to not deliberately make our computers worse.
If they are junior developers working in Java they may just as well build an AbstractFactoryConcurrentSingletonBean because that’s what they learned in school as an LLM would be from training on code it found on the Internet.
If you don't like the results or the process, you have to switch targets or add new intermediates. For example instead of doing description -> implementation, do description -> spec -> plan -> implementation
The more I use LLMs, the more I find this true. Haskell made me think for minutes before writing one line of code. Result? I stopped using Haskell and went back to Python because with Py I can "think while I code". The separation of thinking|coding phases in Haskell is what my lazy mind didn't want to tolerate.
Same goes with LLMs. I want the model to "get" what I mean but often times (esp. with Codex) I must be very specific about the project scope and spec. Codex doesn't let me "think while I vibe", because every change is costly and you'd better have a good recovery plan (git?) when Codex goes stray.
This is technically true. But unimportant. When I write code in a higher level language and it gets compiled to machine code, ultimately I am testing statically generated code for correctness. I don’t care what type of weird tricks the compiler did for optimizations.
How is that any different than when someone is testing LLM generated C code? I’m still testing C code that isn’t going to magically be changed by the LLM without my intervention anymore than my C code is going to be changed without my recompiling it.
On this latest project I was on, the Python generated code by Codex was “correct” with the happy path. But there were subtle bugs in the distributed locking mechanics and some other concurrency controls I specified. Ironically, those were both caught by throwing the code in ChatGPT in thinking mode.
No one is using an LLM to compute is a number even or odd at runtime.
The same thing happens in JavaScript. I debug it using a Javascript debugger, not with gdb. Even when using bash script, you don’t debug it by going into the programs source code, you just consult the man pages.
When using LLM, I would expect not to go and verify the code to see if it actually correct semantically.
Like I said above, I do know to watch out for implementations that “Work on my Machine” but don’t work at scale or involve concurrency. But I have had to check for the same issues when I delegate work to more junior developers.
This is not meant to be an insult toward you. But my not doing front end development for well over a decade, a front end developer might as well be a “human LLM” to me. I’m going to give you the business requirements and constraints and you are going to come back with a website. I am just going to check it meets the business requirements and not tell you the how. I’m definitely not going to look at the code.
I just had a web project I had to modify for a new project, I used Codex and didn’t look at a line of code. Yeah I know JavaScript. But I have no idea whether the initial developer who worked on on another project I led or whether the Codex changes were idiomatic. I know the developer and Codex met my functional requirements.
you might not, but plenty of others do. on the jvm for example, anyone building a performance sensitive application has to care about what the compiler emits + how the jit behaves. simple things like accidental boxing, megamorphic call preventing inlining, etc. have massive effects.
i've spent many hours benchmarking, inspecting in jitwatch, etc.
Yes I know every millisecond a company like Google can shave off, is multiplied by billions of transactions a day and can save real money on infrastructure. But even at a second tier company like Salesforce, it probably doesn’t matter
Over the past decade, part of my job has been to design systems, talk to “stakeholders” and delegate some work and do some myself. I’m neither a web developer nor a mobile developer.
I don’t look at a line of code for those types of implementations. I do make sure they work. From my perspective, those that I delegated to might as well be “human LLMs”.
But even with C, it’s still not completely deterministic with out of order and predictive branching, cache hits vs misses etc. Didn’t exactly this cause some of the worse processor level security issues we had seen in years?
You could use a badly designed antenna with a horrible VSWR at the end of a coax, and effectively communicate with some portion of the world, by using a tuner, which helps cover up the inefficiencies involved. However, doing so loses signal, in both directions. You can add amplification at the antenna for receive (a pre-amp) and transmit with more power, but eventually the coax will break down, possibly well before the legal limit.
It is far better to use a well designed antenna and matching system at the feed point. It maximizes signal transmission in both directions, by reducing losses as much as possible.
--
A compiler matches our cognitive impedance to that of the computer. We don't handle generating opcodes and instruction addresses manually very well. I don't see how an LLM is going to do that any better. Compilers, on the other hand, do it reliably, and very efficiently.
The best cognitive impedance matches happened a while ago, when Visual Basic 6 and Delphi for Windows first came out. You might think LLMs make it easier that that, but you'd be mistaken, for any problem of sufficient complexity.
One current idea of mine, is to iteratively make things more and more specific, this is the approach I take with psuedocode-expander ([0]) and has proven generally useful. I think there's a lot of value in the LLM instead of one shot generating something linearly, building from the top down with human feedback, for instance. I give a lot more examples on the repo for this project, and encourage any feedback or thoughts on LLM driven code generation in a more sustainable then vibe-coding way.
[0]: https://github.com/explosion-Scratch/psuedocode-expander/
Well, you can always set temperature to 0, but that doesn't remove hallucinations.
As for the argument that modern compilers also have nondeterminism in places is still comparing apples to oranges.
LLM's indeterminism means it can always choose from the entire token space. Including junk.
A compiler's nondeterminism is to, possibly randomly, choose from one of a set of VALID solutions. Whatever the nondeterministic outcome is, you are guaranteed a valid solution.
This is why I think the better goal is an abstraction layer that differentiates human decisions from default (LLM) decisions. A sweeping "compiler" locks humans out of the decision making process.
This could be a good way to learn how robust your tests are, and also what accidental complexity could be removed by doing a rewrite. But I doubt that the results would be so good that you could ask a coding agent to regenerate the source code all the time, like we do for compilers and object code.
For context, my initial implementation went through the official AWS open source process (no longer there) five years ago and I’m still getting occasional emails and LinkedIn Messages because it’s one of the best ways to solve the problem that is publicly available - the last couple of times, I basically gave the person the instructions I gave ChatGPT (since I couldn’t give them the code) and told them to have it regenerate the code in Python and it would do much better than what I wrote when I didn’t know the service as well as I do now, and the service has more features that you have to be concerned about
But for reference, we don't (usually) care which register three compiler uses for which variable, we just care that it works, with no bugs. If the non-dertetminism of LLMs mean the variable is called file, filename, or fileName, file_name, and breaking with convention, why do we care? At the level Claude let's us work with code now, it's immaterial.
Compilation isn't stable. If you clear caches and recompile, you don't get a bit-for-bit exact copy, especially on today's multi-core processors, without doing extra work to get there.
One of the first things I tried to have an llm do is transpile. These days that works really well. You find an interesting project in python, i'm a js guy, boom js version. Very helpful.
You see a business you like, boom competing business.
These are going to turn into business factories.
Anthropic has a business factory. They can make new businesses. Why do they need to sell that at all once it works?
We're focusing on a compiler implementation. Classic engineering mindset. We focus on the neat things that entertain us. But the real story is what these models will actually be doing to create value.
It’s guessing using the entire sum total of its ingested knowledge, and, it’s reasonable to assume frontier labs are investing heavily into creating and purchasing synthetic & higher quality data.
Judgment is a byproduct of having seen a great many things, only some of which work, and being able to apply that in context.
For most purposes (granted not all) it won’t matter which of the many possible programs you get - as long as it’s usable and does the task, it’ll be fine.
There are people playing around with straight machine code generation, or integrating ML into the optimisation backend, finally compiling via a translation to an existing language is already a given in vibe coding with agents.
Speaking of which, using agentic runtimes is hardly any different from writing programs, there are some instructions which then get executed just like any other applications, and if it gets compiled before execution or plainly interpreted, becomes a runtime implementation detail.
Are we there yet without hallucinations?
Not yet, however the box is already open, and there are enough people trying to make it happen.
I think it’s more productive to chart all of these systems, LLMs included, on a line of abstraction leakiness. Even disregarding their stochastic nature, I think they’re a much too leaky abstraction to find any use in compilers. There’s a giant mismatch that I think is too big to reconcile.
The obvious has been stated.
LLMs are not designed for that.
But the determinism/non-determinism axis isn't the core issue here. The issue is that they are trained by gradient descent which produces instability/unpredictability in its output. I can give it a set of rules and a broad collection of examples in its context window. How often it will correctly apply the supplied rules to the input stream is entirely unpredictable. LLMs are fundamentally unpredictable as a computing paradigm. LLMs training process is stochastic, though I hesitate to call them "fundamentally stochastic".
You cannot formally verifiy prose or the text that LLMs generates when attempting to compare what a compiler does. So even in this sense that is completely false.
No-one can guarrantee that the outputs will be 100% to what the instructions you are giving to the LLM, which is why you do not trust it. As long as it is made up of artificial neurons that predict the next token, it is fundamentally a stochastic model and unpredictable.
One can maliciously craft an input to mess up the network to get the LLM to produce a different output or outright garbage.
Compilers have reproducable builds and formal verification of their functionality. No such thing with LLMs exist. Thus, comparing LLMs to a compiler and suggesting that LLMs are 'fundamentally deterministic' or is even more than a compiler is completely absurd.
By even doing that, the result of the outputs are useless anyway. So this really does not help your point at all. So therefore:
> You're just using words incorrectly. Deterministic means repeatable. That's it. Predictable, verifiable, etc are tangential to deterministic.
There is nothing deteministic or predictable about an LLM even when you compare it to a compiler, unless you can guarrantee that the individual neurons through inference give a predictable output which would be useful enough for being a drop-in compiler replacement.
[0] https://152334h.github.io/blog/non-determinism-in-gpt-4/
[1] https://arxiv.org/pdf/2506.09501
[2] https://thinkingmachines.ai/blog/defeating-nondeterminism-in...
It is directly relevant and supports my whole point which just debunked your assertions on LLMs being ‘deterministic’ which doesn’t exist at a fundamental sense which you can’t guarantee that the behaviour and even the outputs will be the same.
> The core algorithm is deterministic. And you're still conflating deterministic and predictable.
The entire LLM is still non-deterministic and it is still considered to be unpredictable even if you take that to account.
> It's strange to have such disregard for the meaning of words and their correct usage.
Nope. Not only you have shown absolutely zero sources at all to prove the deterministic nature of LLMs to where it can function as a “compiler”, you ultimately conceded by agreeing with the linked paper(s) recognising that LLMs still do not have deterministic or predictable properties at all; even if you tweak the temp, parameters, etc.
Therefore, once again LLMs are NOT compilers as even feeding them adversarial inputs can mess up the entire network up to become useless.
> As a consequence, the model is no longer deterministic at the sequence-level, but only at the batch-level
therefore they are deterministic when the batch size is 1
Your second source lists a large number of ways how to make LLMs determnistic. The title of your third source is "Defeating Nondeterminism in LLM Inference" which also means that they can be made deterministic.
Every single one of your sources proves you wrong, so no more sources need to be cited.
This is like saying: "C++ is 'safe' when you turn off all the default features and if you know what you are doing", but up to the point where it becomes absolutely useless, and it's still not safe.
The language is still fundamentally memory unsafe, just like how LLMs are fundamentally deep neural networks which that comes with downsides such as being unpredictable blackboxes which carry lots of non-determinism of outputs with that.
> Your second source lists a large number of ways how to make LLMs determnistic. The title of your third source is "Defeating Nondeterminism in LLM Inference" which also means that they can be made deterministic.
That is the point: "When", "Can be made", "how to make LLMs determnistic".
It just tells you something about why both papers recognise the problem of non-determinism in LLMs which makes my whole point even more valid which is why I linked those papers.
Those papers have highlighted the fundamental nature of these LLMs right at the start of the paper.
> Every single one of your sources proves you wrong, so no more sources need to be cited.
Nope, it is quite the opposite.
Note that I never defended using LLMs as a compiler. In fact I argued it would be inappropriate. I simply disagreed that the reason is because they are non-deterministic. If you weren't conflating the meaning of deterministic and predictable, you wouldn't keep misreading me.
Its software. Without an external randomness source, its 100% deterministic excluding impacts of hardware glitches. This...isn’t debatable. You can make it seem non-deterministic by concealing inputs (e.g., when batching multiple requests, any given request is “nondeterministic” when viewed in isolation in many frameworks because batches use shared state and aren’t isolated), but even then it is still deterministic you are just choosing to look at an incomplete set of the inputs that determine the output.
I don't think anyone would even go as far as to include all deep neural networks which are indeed a large scale collection of neural networks as being "100% deterministic"; regardless of their architecture. Not even you yourself and I can explain transparently why it sometimes works or doesn't work the way it does especially with any inputs. (Which adverse inputs can really mess up the model).
But first of all, the entire sentence that you should be quoting for complete context is:
>> You have not shown how a large scale collection of neural networks irrespective of their architecture is more deterministic when compared to a 'compiler' and only repeating a known misconception of tweaking the temperature to 0 which does not bring the determinism you claim it brings with LLMs [0] [1] [2], otherwise you would not have this problem in the first place.
So given this "100% determinism" you just said, surely that means that LLMs can replace a traditional compiler which needs this said determinism, since that LLMs are so useful for such a use case in production?
Then, as we practically test this, all of this quickly falls into my secondary point:
>> By even doing that, the result of the outputs are useless anyway. So this really does not help your point at all.
Again, there is just no point with repeating such myths from AI boosters that deep neural networks like LLMs are '100% deterministic'; even with temp=0 tweaks and in the practical sense.
Any other position is pure magical thinking. Adding more linear algebra never gets you out of 100% determinism. It gets you more complexity, and potentially makes systems chaotic, but chaotic systems are still deterministic, just highly sensitive to small input variations.
I think this is an interesting development, because we (linguists and logicians in particular) have spent a long time developing a highly specified language that leaves no room for ambiguity. One could say that natural language was considered deficient – and now we are moving in the exact opposite direction.
0. "Make something cool out of this insane amount of energy." (temp: 10^42 Kelvin)
1. He slept for a while.
2. Datacenter exploded His realm.
3. ~380 000 years passed and fiat lux.
4. ~13 billions years passed and here we are.
5. JMP 0.This is not really a point about whether LLMs can currently be used as English compilers, but more questioning whether determinism of the final machine code output is a critical property of a build system.
"This gets to my core point. What changes with LLMs isn’t primarily nondeterminism, unpredictability, or hallucination. It’s that the programming interface is functionally underspecified by default."
>From one gut feeling I derive much consolation: I suspect that machines to be programmed in our native tongues —be it Dutch, English, American, French, German, or Swahili— are as damned difficult to make as they would be to use.
yet nobody complained about this
in fact engineers appreciate that, "we are not replaceable code monkeys cogs in the machine as management would like"
You could say that all those things are ultimately dependent on an input and given the same input will always finish with the same result. And that would be a fair assertion. But once you start thinking that way then even LLMs are 100% deterministic. Fundamentally, computers can only operate deterministically. We only sometimes think of them as being non-deterministic because we are willing to ignore some inputs.
This feels like the same debate assembly programmers had about C in the 60s. "You don’t understand what the compiler is doing, therefore it’s dangerous". Eventually we realised the important thing isn’t how the code was authored but whether the behaviour is correct, testable, and maintainable.
If code generated by an LLM:
- passes a real test suite (not toy tests),
- meets performance/security constraints,
- goes through review like any other change,
The real risk isn’t "LLMs as compilers", it’s letting changes bypass verification and ownership. We solved that with C, with large dependency trees, with codegen tools. Same playbook applies here.
If you give expected input and get expected output, why does it matter how the code was written?
> passes a real test suite (not toy tests)
“not toy tests” is doing a lot of heavy lifting here. Like an immeasurable amount of lifting.
Why? Because new languages have an IR in their compilation path?
Can you formally verify prose?
> But yeah, they probably don't fit the bill of English based code to machine code
Which is why LLMs cannot be compilers that transform code to machine code.
Using LLMs to do something like what a compiler can already do is also modelling LLMs as infinite rather than finite. In fact in this particular situation not only are they finite, they're grotesquely finite, in particular, they are expensive. For example, there is no world where we just replace our entire infrastructure from top to bottom with LLMs. To see that, compare the computational effort of adding 10 8-digit numbers with an LLM versus a CPU. Or, if you prefer something a bit less slanted, the computational costs of serving a single simple HTTP request with modern systems versus an LLM. The numbers run something like LLMs being trillions of times more expensive, as an opening bid, and if the AIs continue to get more expensive it can get even worse than that.
For similar reasons, using LLMs as a compiler is very unlikely to ever produce anything even remotely resembling a payback versus the cost of doing so. Let the AI improve the compiler instead. (In another couple of years. I suspect today's AIs would find it virtually impossible to significatly improve an already-optimized compiler today.)
Moreover, remember, oh, maybe two years back when it was all the rage to have AIs be able to explain why they gave the answer they did? Yeah, I know, in the frenzied greed to be the one to grab the money on the table, this has sort of fallen by the wayside, but code is already the ultimate example of that. We ask the LLM to do things, it produces code we can examine, and the LLM session then dies away leaving only the code. This is a good thing. This means we can still examine what the resulting system is doing. In a lot of ways we hardly even care what the LLM was "thinking" or "intending", we end up with a fantastically auditable artifact. Even if you are not convinced of the utility of a human examining it, it is also an artifact that the next AI will spend less of its finite resources simply trying to understand and have more left over to actually do the work.
We may find that we want different programming languages for AIs. Personally I think we should always try to retain that ability for humans to follow it, even if we build something like that. We've already put the effort into building AIs that produce human-legible code and I think it's probably not that great a penalty in the long run to retain that. At the moment it is hard to even guess what such a thing would look like, though, as the AIs are advancing far faster than anyone (or any AI) could produce, test, prove out, and deploy such a language, against the advantage of other AIs simply getting better at working with the existing coding systems.
Stop this. This is such a stupid way way of describing mistakes from AI. Please try to use the confusion matrix or any other way. If you're going to try and make arguments, it's hard to take them seriously if you keep regurgitating that LLM's hallucinate. It's not a well defined definition so if you continually make this your core argument, it becomes disingenuous.
You can see it clearly if you just translate the article's expensive vocabulary into plain English. When the author writes, 'When you hand-build, the space of possibilities is explored through design decisions you’re forced to confront,' they are just saying, 'When you write code yourself, you have to choose how to write it.' When they claim, 'contextuality is dominated by functional correctness,' they just mean, 'Usually, we just care if the code works.' When they warn about 'inviting us to outsource functional precision itself,' they really mean, 'LLMs let you be lazy.' And finaly, 'strengthening the will to specify,' is just a dramatic way of saying, 'We need to write better requirements.' It is obscurantism plain and simple. using complexity to hide the fact that the insight is trivial.
But that is just an estethical problem to me. Worse. The argument collapses entirely when you look at the logical leap between the premises.
The author basically argues that because Natural Language is vague, engineers will inevitably stop caring about the details and just accept whatever reasonable output the AI gives. This is pure armchair psychology. It assumes that just because the tool allows for vagueness, professionals will suddenly abandon the concept of truth or functional requirements. That is a massive, unsubstantiated jump.
If we use fuzzy matching to find contacts on our phones all the time. Just because the search algorithm is imprecise doesn't mean we stop caring if we call the right person. We don't say, 'Well, the fuzzy match gave me Bob instead of Bill, I guess I'll just talk to Bob now.' The hard constraint, the functional requirement of talking to the specific person you need, remains absolute. Similarly, in software, the code either compiles and passes the tests, or it doesn't. The medium of creation might be fuzzy, but the execution environment is binary. We aren't going to drift into accepting broken banking software just because the prompt was in English.
This entire essay feels like those social psychology types that now have been thoroughly been discredited by the replication crisis in psychology. The ones who are where concerned with dazzling people with verbal skills than with being right. It is unnecessarily complex, relying on projection of dreamt up concepts and behavior, rather than observation. THIS tries to sound profound by turning a technical discussion into a philosophical crisis, but underneath the word salad, it is not just shallow, it is wrong.