>>108069225
But /g/ told me AI is just autocomplete slop generator. How can this be?

>>108069225

okay now we're talking.
next step making an os.

>>108069225
How does it compare to GCC, Clang and MSVC?

File: 20260129_074618.jpg (105 KB, 1440x1440)

105 KB JPG

>>108069225
Some? I see a big fucking caveat that they wasted $20k compute time on something that is completely useless in practice. If it's supposedly so good, why not put it working on a problem that needed solving?

>>108069292
This is a much more reasonable take than the cursor-browser thing. A few things that make it pretty impressive:

> This was a clean-room implementation (Claude did not have internet access at any point during its development); it depends only on the Rust standard library. The 100,000-line compiler can build Linux 6.9 on x86, ARM, and RISC-V. It can also compile QEMU, FFmpeg, SQlite, postgres, redis

> I started by drafting what I wanted: a from-scratch optimizing compiler with no dependencies, GCC-compatible, able to compile the Linux kernel, and designed to support multiple backends. While I specified some aspects of the design (e.g., that it should have an SSA IR to enable multiple optimization passes) I did not go into any detail on how to do so.

> Previous Opus 4 models were barely capable of producing a functional compiler. Opus 4.5 was the first to cross a threshold that allowed it to produce a functional compiler which could pass large test suites, but it was still incapable of compiling any real large projects.

And the very open points about limitations (and hacks, as cc loves hacks):

> It lacks the 16-bit x86 compiler that is necessary to boot [...] Opus was unable to implement a 16-bit x86 code generator needed to boot into 16-bit real mode. While the compiler can output correct 16-bit x86 via the 66/67 opcode prefixes, the resulting compiled output is over 60kb, far exceeding the 32k code limit enforced by Linux. Instead, Claude simply cheats here and calls out to GCC for this phase

> It does not have its own assembler and linker;

> Even with all optimizations enabled, it outputs less efficient code than GCC with all optimizations disabled.

Ending with a very down to earth take:

> The resulting compiler has nearly reached the limits of Opus’s abilities. I tried (hard!) to fix several of the above limitations but wasn’t fully successful. New features and bugfixes frequently broke existing functionality.

>>108069292
from the post:
>The compiler, however, is not without limitations. These include:
>It lacks the 16-bit x86 compiler that is necessary to boot Linux out of real mode. For this, it calls out to GCC (the x86_32 and x86_64 compilers are its own).
>It does not have its own assembler and linker; these are the very last bits that Claude started automating and are still somewhat buggy. The demo video was produced with a GCC assembler and linker.
>The compiler successfully builds many projects, but not all. It's not yet a drop-in replacement for a real compiler.
>The generated code is not very efficient. Even with all optimizations enabled, it outputs less efficient code than GCC with all optimizations disabled.
>The Rust code quality is reasonable, but is nowhere near the quality of what an expert Rust programmer might produce.

I definitely expect it to be a shitty implementation in terms of quality of code generated (probably also compilation speed). Likely has MANY codegen bugs lurking in there too, but the sheer accomplishment here is insane.

If this rate of progress is sustained, what will the best frontier model be able to do in a year? How long before it writes a better compiler backend than LLVM that produces better code and has fewer bugs and compiles faster? I'm not sure that will happen but it doesn't seem as unrealistic of a feat as it did just 1 year ago.

>>108069329
>How long before it writes a better compiler backend than LLVM that produces better code and has fewer bugs and compiles faster?
Once the learn step gets measured against uop benchmarks and has must meet robust unit tests

>>108069307
>>108069329
>Rust
This takes a century to compile though. Why didn't they pick something like Fil-C if they're obsessed with memory safety?

>$20k api costs

>>108069225
>training data consists of c compilers
>generate compiler that performs worse than those it was trained on
grim

>nothing happened
>nothing will happen
the only happening is the Epstein files.

>>108069351
I don't get why u would care about memory safety for a compiler either, you can just malloc everything (slow) and it throws away the memory when it's done compiling anyways, it's a run once type of program. There's no WASM target and I don't know if it compiles for WASM either I would've seen that rather

>>108069225
>A human guided some of this process by writing test cases that Claude was told to pass
>For example, near the end of the project, Claude started to frequently break existing functionality each time it implemented a new feature. To address this, I built a continuous integration pipeline and implemented stricter enforcement that allowed Claude to better test its work so that new commits can’t break existing code.
Why those niggers hide the fact that even a gazilion agents can't really do anything on their own? $20K for a compiler that "works" but needs to rely on GCC for the linking and assembling? And produce code that runs slower than gcc -O0, nice! Can we stop wasting everybody's time already?

>>108069450
The steel man case for something like this is that you can pay 1 experienced developer to make some complex project by wrangling $20k worth of semi-competent agents rather than paying a team of developers, who would be more expensive and probably slower.
Yes, that does move the goal post several countries over from the original claim of "AI will replace all human programmers in 2 more weeks."

>>108069370
This.
20K api costs to Claude retrieve what already is in the tranning data... Plus human assistance

>>108069299
lost

it's absolutely fucking over for software 'engineers' in a couple of years

>>108069364
cheap for the amount of work
and getting cheaper every year

>>108069225
Fuck off
https://github.com/anthropics/claudes-c-compiler/issues/1

>>108069609
It's cheap for a compiler that actually works, but absurdly expensive for this quality of a toy project. How much more work would it take to make the compiler output actually decent and usable?

>>108069619
>How much more work would it take to make the compiler output actually decent and usable?
a couple of years of work on the llm ;-)

>>108069609
should've went for cybersec/it

Fuck, I'm genuinely mad at this point.
They're obviously using this shit to scam managers to have another wave of layoffs.
First "browser", now "compiler"

It doesn't work, fuck off anthropic employees
Why the fuck whould you even promote it on /g/?

>>108069307
oh my god
isn't there a possibility it's just translating clang or gcc into rust?

>>108069370
not only that, but it was allowed to verify results by calling gcc

that is a massive crutch and basically turned it to a looping translation task

>>108069225
>>108069307
why can't LLMs simply translate source code into binary programs?

>>108069611
hello world worked on my machine.

>>108069794
who says they cant

>>108069799
well why don't they do that? it would be more interesting than what they did

>>108069797
ye hes a retard probably missing glib, he could've just used musl too

>>108069266
C compiler already exists so it will essentially be autocomplete copy pasting

tell it to invent a new language and write a compiler for it and see what happens retard

>>108069803
easier to verify, more flexible and debug just in case, the same reason most c replacements first transpile to c

>>108069797
>>108069805
nice try anthropic employees, how much do they pay you for? I hope it's at least 1$ per post

>>108069828
>nice try anthropic employees
i wish, do they get unmetered access to claude? hnnnnnnng

>>108069225
Terry Davis can do it for free.

It's all so tiresome.

Use case for using Rust to build C?

>>108069851
there are almost none, the opposite would be funnier using C to compile Rust

File: 1300544820342.jpg (51 KB, 450x337)

51 KB JPG

>>108069609
>it's absolutely fucking over for software 'engineers' in a couple of years

we shall see
for now everyone 1) happily ignores maintenance and 2) happily ignores the fact that without juniors there won't be any seniors
if anything, the CEOs will be replaced faster than senior IT staff, because investors will realize that their trust is better placed in a machine that does not fuck around, sell secrets etc.

>>108069838
no he can't because he's fucking dead

File: Renge Haskell Programming.jpg (960 KB, 1920x2443)

960 KB JPG

>>108069225
It can't do a Haskell compiler

File: shot-2026-02-05_22-37-04.jpg (83 KB, 641x407)

83 KB JPG

>>108069828

did work.

>>108069364
And how much do you reckon would it cost for you to write a comparable C compiler?

File: file.png (210 KB, 1764x824)

210 KB PNG

>>108069611
cope

>>108069307
>clean-room implementation
Except for having an absolute shitton of code already baked in lol.
Makes you wonder why they chose 6.9 (May 2024) instead of .16,.17 or .18??

>>108069923
$5000

>>108069225
>you can pay TWENTY FUCKING THOUSAND DOLLARS to have this ai steal code from github and give you something that almost works but not really
Oh my science, we achieved AGI.

>>108069937
Shhh! You are noticing too much

>>108069225
this is stupid you don't use your training set in the validation run for a reason. Of course it will arrive at a solution it has already consumed every existing c compiler.
If they really wanted to show how good claude is then implement a standalone compiler for zig. Andrew has been blueballing everyone for a decade now.

File: Screenshot from 2026-02-0(...).png (124 KB, 1401x642)

124 KB PNG

>>108069926
Lmao guys just stop at this point please

Baggies gonna bag, it's useless to try to debate the silliness of this supposed achievement.

>>108069225
time to vibeport llvm to rust

>>108069812
Could you do that on your own?

>>108069225
You really be eating shit up from a company that made terminal in React? Are you for real right now?

>>108070041
I can copy paste a C compiler on my own, yes.

>>108070041
With enough time, book reading, and drive I'm sure I could cook something very rudimentary up

(No)

>>108069225
>it runs DOOM
AGI is right here brothers.

can it write a jai compiler?

>>108070005
remove the space from between -I and the path

File: le pepe.jpg (95 KB, 798x798)

95 KB JPG

>>108070041
Just two more trillions bro.

>>108070041
Its okay we still like you more than gpt dont be too upset sweetie.

>>108069225
why aren't these giga niggers doing anything useful? Its always some pr shit thats useless to the public. Do some real work implement some linux drivers get it accepted, work on blender, etc. Anything but meme shit.

File: pepe-dancing.gif (1.61 MB, 452x498)

1.61 MB GIF

>>108069609
Of course, you are very intelligent.
Create your own software development firm, 100% AI.
Sell everything you own and go all in.
You will be the first trillionaire!

I don't remember seeing a biggest scam, ever. Grifters have always existed, but I can't remember they were this fucking ridiculous, ever.
>our magic tech can prove this mathematical problem!
(it literally just looked for the solution and copied it)
>our magic tech can make a full browser!
(it literally won't even compile)
>our magic tech can program a secure messaging app with post encryption security!
(it literally was full of TODOs in every method because it couldn't even check certificates)
>our magic tech can create a compiler that works on its own!
(it literally doesn't)
Time after time, absolutely false and ridiculous claims only to get more money from people who want to believe it, they literally don't care that hundreds of thousands of people can see that every single time nothing works.

>>108069225
So they spent 20k doing something a team of engineers could do for 5?

>>108069307
>Claude did not have internet access at any point during its development
stopped reading as this is not honest and also a lie

File: 1755282045107442.png (18 KB, 527x189)

18 KB PNG

Is this one of you faggots? I definitely saw this username in one of the last threads.

>>108069225
Wow! Are you telling me that a machine has regurgitated gcc, which is free and open source???

>>108069225
RMS. TAD. Their entire claim to fame just $20,000 now
it's so ogre

>technology board
>people here mock and ridicule new technologies
????

>>108070369
yeah like the juicero this is a complicated way to squeeze a bag

>>108070369
>new tech
>just copy pastes old tech from it's training corpus with a 80% accuracy

>technology board
>people here don't even have a juicero
????

>>108070369
It's not a "new technology" it's digital fucking theft

>>108070369
Can this new technology tell me how many r's are in strawberry reliably?

>>108070414
that's lacist

>Rust
no thanks

>>108069225
>Pirate the entire internet, including the C compiler and 401285 derivatives of it.
>Use it to model likelyhoods of words following other words
>Tell it to create a C compiler
>It creates a C compiler

WOW BIG NEWS

>>108069731
Of course it is. You think it's a coincidence that they always ask it to reimplement some high profile program that was definitely in the source code, rather than asking it to write something novel?

>>108069299
What's going on in this picture?

>>108069812
I kinda did. I tried it to build a LISP like language (LISP syntax but with some different ideas) and I asked for the following features:

> Floating point support
> Garbage collector
> Tail call optimization
> Macro support (including quasiquotation)
> Static typing (one of the "different ideas")
> OS integration (I/O and basic networking)
> A standard library including functional data structures (following Okasaki's work) and a small http wrapper

I then asked to translate an existing web project into this language using its idiosyncrasies and design patterns. It managed to translate the whole projects into the new language extending some of the design ideas that I initially suggested (since I expressly said to try overusing the feature set).

The project worked, granted it was a bit slower compared to the original Java implementation, but it worked fine. I also convinced my former employer to run it in production (since it was an non essential program) for a couple of days and it served our users perfectly fine.

I know that nobody here would believe anything that I wrote so far, but these tools are perfectly capable of doing 90% of what a typical developer usually do (ie gluing a framework to a database or to an user).

>>108069225
>$20,000 in cost for AIslop
This is an advertisment to generate profit.

>>108070513
>I know that nobody here would believe anything that I wrote so far, but these tools are perfectly capable of doing 90% of what a typical developer usually do (ie gluing a framework to a database or to an user).
Considering that there are many many lisp interpreters that can be found on github and are by that in the training data of llms, it's pretty believable. Just kinda sad that pretty much takes out the fun of writing one owns language (for which there isn't usually a real reason other than to have fun)

>>108070513
>I know that nobody here would believe anything that I wrote so far
To be fair I don't even think your retarded babble was written by a person.

>>108070513
>>108070544
I was planning to attempt the same thing because logically Lisp makes the most sense for these language models. It's practically English already with parenthesis, and the reduced featureset being just simple concepts should compliment context limits better than say complex languages. Much harder to make syntax errors when everything is the same format.

File: Writing a C Compiler.jpg (13 KB, 170x225)

13 KB JPG

>>108069225
https://nostarch.com/writing-c-compiler
They've bought the book.

>>108070477
RUMBLING, RUMBLING

>>108069225
It‘s over, time to pick a new career. I‘d say one more year until most teams will just be 1-2 people and the rest a bunch of agents.

>>108069812
>C compiler already exists
In Rust? An end-to-end C compiler?

File: SMITHCLONES.jpg (39 KB, 600x338)

39 KB JPG

>>108070858
Oh my Musk, it's just like in my science-fi movies!

>>108069480
>can pay 1 experienced developer to make some complex project by wrangling $20k worth of semi-competent agents rather than paying a team of developers

Can you? For all we know that code is a complete unfixable mess and it's impossible to improve further without rewriting all of it. And that would require a full team of developers...

how will codejeets cope
it's fucking over

>>108070041
Yeah? Writing a basic C compiler for one architecture was literally a one semester collage project. It can't be that hard for real professionals.

And I could just cheat like Claude did by compiling into C and then compiling using gcc.

>>108070939
>saar pls to be redeeming AI Israel phul sapport
>it good saar 2 lakh rupee only

>>108069225
>a language model trained on the source code from C compilers generated a C compiler
holy shit

File: 1739944218986549.png (313 KB, 735x656)

313 KB PNG

>>108069225
>thousands and thousands of posts shilling this shit
>not a single post with an impressive code made using it
Hmm... I think I am starting to notice.

>>108069225
I could do that with $200 of Claude using one agent.

>>108069307
Niggerlicious. 100k LoC for a new compiler only for x86-64 for language like C is too much.

>>108073770
>successfully compiles the Linux kernel, qemu, ffmpeg, sqlite, postgres, and redis
You're an illiterate retard and you'll deserve being completely replaced by AI.

>>108069225
Let me guess, it works the same way as cursor's browser, or doesn't work at all.

File: opusgoatse.png (25 KB, 447x537)

25 KB PNG

>>108069225
>claude opus took hours to copy-paste a c compiler from internet
ftfy

File: 545873932-9a3a08c0-90d1-4(...).png (85 KB, 2228x886)

85 KB PNG

>>108073816
It's even worse. I have to test it by myself but from what I can see it made C 10 times slower.

>>108070369
>new technologies
>new
is "new" with us in the same room right now

I wasn't too surprised when artists were melting down over AI, but the crash out from programmers is much, much, funnier.
Just completely in denial at this point.

>>108069225
20,000 to regurgitate gcc and copy and paste from git. The world is truly bananas.

>>108070975
>by compiling into C and then compiling using gcc
>compiling into C
so it is just 'cat'?

>>108069225
I don't understand what is amazing about this.
It wasted $20k to make C run 10–100 times slower.
I tested it, simple matrix calculation is 12 times slower.
And creating a terrible compiler isn't that hard.
You parse C code and turn it into assembly based on set rules.

File: Grain fed.jpg (93 KB, 500x455)

93 KB JPG

>>108069225
>some caveats exist but this is fucking insane

>>108069292
>can't even build fucking hello world without you manually pointing it to GCC's headers.
>assembly is worse than some toy compilers much less GCC without any optimization enabled
Try harder

>>108074009
Unless you are part of the billionaire/ai owner class I don‘t see how eternal techo serfdom is funny as a future.

File: 1133043955351.jpg (87 KB, 650x426)

87 KB JPG

>>108069225
>a glorified matrix multiplier understood a problem solved and described multiple times over last few decades in hundreds of volumes of digitized technical documentation

unbelievable!
now tell it to implement newest law changes in property tax in an existing system, lol
you'll need MORE developers to fix shit broken by AI, and soon

File: 1770380139144.jpg (6 KB, 225x225)

6 KB JPG

>>108074376
>he isn't a billionaire
Shiggy diggy

>>108074421
>now tell it to implement newest law changes in property tax in an existing system, lol
tricky question, an actual intelligence would abolish it

singularity as close as it gets

File: AGI.png (65 KB, 1666x415)

65 KB PNG

fuck me

>>108069225
>"building" something it was trained on
when will you retards learn that this is no different than querying data from DB, I could google open source C compiler 20 years ago

>>108074363
It can build a Linux kernel. Can you?

How should I evaluate claude for my workflow? It seems to work ok, but I don't really know how much time it saves me.

>>108073681
I wrote a few things with claude but you probably would call it unimpressive.

>>108069329
>How long before it writes a better compiler backend than LLVM that produces better code and has fewer bugs and compiles faster? I'm not sure that will happen but it doesn't seem as unrealistic of a feat as it did just 1 year ago.
This will require a breakthrough of some kind.

File: 1132237741038.jpg (29 KB, 518x669)

29 KB JPG

>>108074684
>I wrote a few things with claude

I beg your fucking pardon?
you didn't write anything, you asked your coworker sitting next to you to write it for you
and he/it didn't even write it, he pirated it from existing resources

>>108070463
>>108070381
You have access to the entire internet too. Go make a C compiler on your own, you can even copy paste all the code you want, you just can't directly use any existing compilers.
>>108070402
>theft
I would download a car.

>>108074764
>You have access to the entire internet too. Go make a C compiler on your own

1) there's no point, no one needs a new C compiler, it's something that is already solved
2) I don't have over 9000 state-sponsored GPUs linked to my brain

>>108070544
>>108070701
Yeah it's pretty sad. I didn't write that in an optimistic way. In fact, there's little to no reason to build anything from scratch these days. One of the reason of why I started programming decades ago was because I could build whatever I wanted from scratch, that's how I and many others have learned. I wouldn't blame youngsters from not pursuing that field when there's something faster and more accurate than you that can even tackle hardcore projects. I witnessed many changes over the years, but this one is just something else.

File: blame__sanakan_by_darkpul(...).jpg (90 KB, 1024x542)

90 KB JPG

>>108074783
>but this one is just something else.
AGI means we go extinct by 2030. you and everyone you know will be physically murdered by ai.

It does not suck my dick and it does not get clapped with the value it takes to run if it makes egregious mistakes.
Why should I care about it?

It doesn't rim my ass or can get impaled by my dick.
No real women does either, but that's why they are not staying under my roof.

>>108074684
Proof next thread?

>>108070369
Webshitters are seething since they've been fired in drove

>>108069266
It is.

>>108074576
Perfectly valid C, as written in the standard. This warning is optional and is a GCC extension.

File: AI scaling fallacy.jpg (452 KB, 1456x819)

452 KB JPG

>>108074805
>>108074783
It's ok, you can rest easy because AGI is literally impossible with transforming LLMs. It's something created by retards like Altman and musk who are attempting to trick investors into wasting their money.

>>108074779
Can you even compile Linux with an existing C compiler?

>>108074815
>Webshitters are seething since they've been fired in drove
saaar we master poopoo saaaaaaaar we generate new web standard saaaar

>>108069225
C compiler was made so many times that it's basically copy paste without attribution.

>>108074849
How do you think people compile the kernel?

>>108074849
I did ~20 years ago with minix - we had workshops about changing and recompiling the kernel
and it was just one assignment out of many

>>108074859
It's not a question of whether SOME people can do that, but whether you can do it.
>>108074870
I think we're the minority

>>108074667
It can't without GCC holding its hand, which Anthropic admited to it

>>108069275
Unironically might be possible soon.

AI can already solve complex individual problems 20 times faster than a human (with some caveats). You just need to structure each individual piece of work in a hierarchy of multiple levels in a way that enables it to put it all together and has protocols for troubleshooting. The typical caveats can be worked around by AIs prompting each other in more chaotic ways until the problem gets solved. Basically, sometimes you have to tardwrangle an AI as a human, but in the long run, you should be able to automate all that.

In a press of a button you can have a swarm program an operating system in a month for far cheaper than with real humans.

Though I guess the biggest issue right now is that the vision and navigation capabilities of agents are too dumb to test evaluate a software with human sensibilities to be able to iterate on it, which for now requires human feedback.

>>108074988
>I think we're the minority
well OF COURSE we're minority, or are you suggesting that just everyone should be able to write a compiler or any program, really?

>>108074844
cope

File: 1745838061247393.png (583 KB, 1082x1474)

583 KB PNG

>>108073681
https://greasyfork.org/en/scripts/564557-mal-alternative-statistics

I made this userscript for MAL using Gemini 3, I just told it to get the data from the Stats page and calculate the stuff the way I wanted (verified same result by asking Grok to calculate same things manually), told it where to put the box, told it to use placeholder numbers instead of loading screen so the layout stays stable, add hoverable info boxes, change layout a bit. I haven't written a line of Javascript in my life, though I did make games as a kid long time ago.

Though this is a pretty simple script. Probably breaks down if you try to do something more complicated. Most of the changes were made with Thinking after running out of the free Pro uses, and I don't think it made a single mistake. As long as you can clearly define what you want, it does it.

File: carbon.png (158 KB, 856x1154)

158 KB PNG

>>108074844
right

You only need to remember the 4 U's when dealing with "AI": Unmaintainable, Unreliable, unoriginal, Unusable.
No need to pay 20k on API, it would be faster and cheaper to ask the LLM where is(one) of the source they got trained and copy yourself.
The idea that some "people(jeets)" pay and wait till the shatbot remember the whole thing (+ bugs) is just sad

>>108075184
>I made this userscript for MAL using Gemini 3

you didn't make shit

>>108075204
Pardon my inaccuracy, I had an AI make this script for me. Which was the point. You don't need to code.

>>108075214
then I guess you're fine with ai hallucinating a script that will publish your credentials, write raunchy posts as you etc.
if you won't code, you'll lose the ability to understand the code. will you 100% believe LLM when it tells you what it does?

>>108075241
I did read through it, seeing mostly math and layout related stuff, but with how well it was following the instructions decided that it's unlikely to do anything dangerous by asking it to reading a webpage and print stuff on the page. Of course if you use things as a part of a bigger project or are using dangerous tools the risks are higher.

I also once asked it to make a .vbs script to use mkvmerge to remux some audio to a different anime release, told it to use the files within the folder and put the results in a subfolder. Despite not asking it to delete anything, I did read the code more carefully since a .vbs script could potentially do big damage. Yeah I heard about the case of AI wiping someone's hard drive. Writing code takes and hour, skimming through an AI code takes a minute.

I don't code nowadays but I'm happy I have an AI make actual useful things happen. Maybe it's not as usable for big projects yet, but it's only a matter of time.

>>108074813
I have to create a new GitHub account. But the projects really are nothing special, would be cool if someone posted theirs.

>>108075325
>skimming through an AI code takes a minute.

for now. it will take longer and longer when you stop coding, you'll start forgetting things.

>>108075339
But desu it was still a good help, e.g. for the Clion plugin I wrote. They are written in Java or Kotlin and the last time I used Java was probably 15 years ago. If I could write it in C++ or C# or Python it would've been easy, but I just asked Claude to do it. I still understand what it does but it probably saved me a few hours of pointless syntax errors.

>>108074764
Sure, I will. Just give me 10T $ too or however bloated the bubble is right now

>>108075359
but I stopped 10 years ago...

>>108069923
My time is worthless

>>108069225
There are tons of hobby projects out there.
Someone should ask it to implement something truly innovative for which there is no code out there (e.g., rocket stabilization for reusable rockets, which I'd guess there is far less open content) and see how it does.

>>108075379
heh. guess the script was simple enough.
think about future "juniors" though, trying to understand ai written code without having an iota of experience of current programmers

>>108075393
AI will be horrible for society, but that shouldn't be confused with it being bad at generating at least simple code.

>>108069225
sounds stupid. maybe go at it in stages next time, start with something like tiny c. maybe spend less than $20k.

>>108075393
Just have an AI read and explain the code. It worked at least for my script. Basically, we have fully translated programming into natural language. Natural language in, natural language out.
But yeah I can see it become slightly disastrous when the slop accumulates beyond certain point. Unless it's somehow able to self-correct.

>>108075463
if one model makes a mistake by introducing a horrible bug, why exactly would you trust another model to notice the bug? it may work, but the risk is still there

I think it's easier reviewing people, because they produce LESS code, so it's easier to check it out

>>108069225
instead of using 500 parallel claude agents, how about these AI companies allow for more than 50k token in "memory", aka forgetting shit after 10 minutes

>>108069225
>but this is fucking insane
I don't understand why this is so insane. C compilers are open source and these LLMs were trained on that source code certainly. Why is it impressive that the text reproduction machine has reproduced text it was trained on? Isn't this LITERALLY what it is designed to do? Reproduce parts of its training based on input telling it to do so?

It feels to me that you (and everyone else doing this) falsely equivalate the complexity and difficulty in writing a particular text from nothing (whether it is source code or anything else), including the reasoning required to come up with it in the first place, to the complexity of making a program which reproduces parts of text it was already given. If you are merely reproducing something (or just doing so poorly, even) then you are going to see far less complexity than actually creating it the first time around.

File: HAaalRiXoAEizOB.jpg (49 KB, 1280x720)

49 KB JPG

>>108075482
4.6 has initial/experimental support for 1m context length and long-context performance has skyrocketed in the past couple of months
previously they would mostly shit the bed at around half their context windows but something happened and both gpt and opus had huge performance increases

File: sam-terminator-fixed.png (1.22 MB, 1024x1024)

1.22 MB PNG

>>108069225
>i made a compiler, with fully autonomous agents
>BUT
>no assembler, no linker (the easiest part. its an a = b substitution, basically)
>compiler is a buggy mess, and when it works, the code is worse than unoptimized gcc
>needs to use human made tests

how is this insane? this is on par with the expectations
its fukken ai slop

cool just recreating something it's already trained on... why are people impressed with this?

>>108074838
Nah

File: Screenshot from 2026-02-0(...).png (60 KB, 1122x539)

60 KB PNG

>>108075573
theyre not. its a pump and dump scheme
the only ones who have a shot at shatbots are jewgle, msft and Xai because they have other sources of revenue

>>108075573
wait till you find out about mathematics

>>108075463
Natural language is inherently not precise enough. That was the beauty of programming, precision and determinism, both are gone.

ai is cute
i want to have a robot gf

>coders thought they wouldn't get replaced and shat on drawfags for 4 years straight
>they are gonna get replaced even harder than artists
Brutal.

>>108075746
the article says the opposite, cargo cultist retard

File: file.png (72 KB, 1023x629)

72 KB PNG

>>108075660
buy an ad, bro

>>108075764
Shut up jeetoid

File: 1757456877905337.png (742 KB, 795x795)

742 KB PNG

>>108070477
shonen anime

>>108075769
>buy an ad
>posts ad
how tf does that relate to the curve and the story it tells?

File: ai-cultists-brown.jpg (149 KB, 1043x300)

149 KB JPG

>>108075770
>calls you a jeet

File: 1765660545162670.png (1.18 MB, 991x4525)

1.18 MB PNG

>>108075795
>one person false flags on slash gee slash
>that must mean that anyone who uses AI coding tools is Indian

>>108075785
the story that curve tells is that someone has decided to make up a graph and try to scam people into gambling on charts that are controlled by the website owner

>>108075816
this compiler is a buggy piece of shit
and half's missing (not even accounting for optimization, or the various comilation options)
cope harder
agi in 2mw faggot

>>108069225
>>108069307
so it basically glued together existing technologies and created something worse.
why do they present it as if it built it itself?

>>108075862
no, it generated some code.
which, expectedly, is a buggy mess and is inferior to gcc -O0

yeah, gcc with no optimizations whatsoever produces better code that this piece of shit
i wonder how thats even possible to be honest

>>108075889 cont
yeah
how can it fuck up a straight up translation of the c code into asm?
this makes no sense to me whatsoever
im at a total loss

File: Roll as Arale by Hideki I(...).jpg (199 KB, 1500x2122)

199 KB JPG

>>108075816
>>108075843
both of you are factually correct and those facts are not mutually incompatable.

>>108069225
saw my code I wrote a year ago and I became depressed, how do I deal with the fact that AI is better than me?

>>108075930 cont of cont
unless yeah
its a pr stunt. its not meant to make sense
its just a wordsalad thats merely supposed to sound impressive
that computes

>>108075482
Because context rot will ruin it for you

>>108075393
I think future juniors won't be reading code that much. It'll be AI dealing with the code. Modern developers don't read any assembly either, we just muck about in our high level languages until shit works.

>>108075889
>it generated some code
great so it generated some useless code that doesn't make any sense in the compilers realm
so there is not difference between this and having a randomize function shitting stuff on top of something

>>108075938
mm
my other interlocutor being sub 80; i remain unconvinced

>>108075977
its not random gibberish either.
if the article is to be believed

because again
-O0 is a straight up translation of the c code into the equivalent in asm
i dont see how you can make that more or less efficient
c is supposed to have an abi
and its not like theres 50 different ways of writing a naive, unoptimized loop afaik

O0 is O0. how can their compiler fuck that up?
genuine question

File: 8cb.jpg (103 KB, 530x836)

103 KB JPG

>>108069225
>LOOK AT THIS THING I HAVE NO IDEA WHAT IT MEANS IT'S INSANE
I support a death penalty for shills and retarded people

>>108075198
>You only need to remember the 4 U's when dealing with "AI": Unmaintainable, Unreliable, unoriginal, Unusable.
That's pretty good. I'll use it in the future.

>>108076005 cont
maybe its that their compiler doesnt use registers at all and everything sits in cache
can any anon enlighten me?
because im not downloading the whole rust environment just to compile this piece of shit
im not sure i even have the space on my pc atm

>>108076005
i didn't say it was random, i just said it was meaningless as if it was random because like you said it doesn't understand cdecl or other routines

>>108076029
ah ok
for my defence, im awake since 17 hours (trying to regulate my sleep patterns, woke up yesterday at midnight)

>>108076047
all good bro
please dream about OP melting in hell flames

>>108070870
There's several from just a quick search.
https://github.com/PhilippRados/wrecc/tree/master
https://github.com/ClementTsang/rustcc
https://github.com/maekawatoshiki/rucc

>>108076070
i sure will

so ...
it was iteratively altered until it accepted all the legal codebase collected for this specific purpose?
it's like an evolutionary algorithm approximating a solution, in this case until it's good enough?
how does it know what to reject, then? in fact how was it told to create a C compiler? just in those words, or was it explicitly supplied with the standard and textbooks?
anyway, I predict it will accept nonsensical programs galore, since it was told to accommodate lots of constructs but never shown wrong code.

File: 1740979621913095.png (397 KB, 628x632)

397 KB PNG

I just made the mistake of using my personal email for claude account and then using that to try refactoring an open source project of my alter ego. So now I'll just wait in horror for anthropic data leaks so that people can put together 1+1 as for who is the owner of that repository. I'm so fucking dumb.

>>108076265
Jesus christ this is slow. How do vibe coders cope? Like claude is taking minutes for deciding what to do.

>>108075977
allegedly a lot of apps compiled with it passed the tests supplied with those apps, like sqlite etc.

>>108076108
>unfinished, can't compile the kernel
>As of now, rustcc only supports variables of type int.
>uses llvm as backend
However poorly the AI is doing, it's still doing more than those.

so the compiler outputs assembly text for a number of different architectures? it's hard to make out technical details from that article.

>>108076501
Sounds like a goalpost move to me.

>>108076549
The original argument was that it's copy-pasting an existing compiler >>108069812
Using LLVM refutes that it's end-to-end.
My replies where how about it couldn't have been copying any of these, because they're not sufficiently functional to compile and run the kernel.

>>108069225
$20k to make something useless is wasting $20k
I wonder how much it would cost to make something that's 100% complete, not this "halfway there" shit they keep marketing with all the time

presenting a prototype of something was already a thing before LLMs, everyone knows the last 20% is the hardest part. also making something that can be changed later is important, which doesn't seem to be the case her ate all

>>108076588
it's a 20k marketing spend
not even a rounding error compared to their superbowl ads

>>108076575
>Using LLVM
One does.

>it couldn't have been copying any of these
Not the toy rust ones obviously but it can translate gcc/clang code to rust retard

>>108076265
>>108076295
dumb tranime addicted NEET with unwarranted self importance

File: 1748081083036.jpg (294 KB, 1220x906)

294 KB JPG

>>108076501
>can't compile hello world

Luddite cope thread
AI GODS WON

File: 1763144726078266.png (262 KB, 680x680)

262 KB PNG

>>108076735
That's not a healthy obsession, anon.

I just want to program, preferably something useful. Please don't make me just fix jeet prompt AI code.

>>108073863
That's pretty fucked

>>108074341
>>108073863
It's indeed pretty bad

and codex 5.3 is even better than opus 4.6
codejeets on suicide watch

>>108075197
Straight from 'The C Programming Language" Second Edition page 4.

>>108075035
>AI can already solve complex individual problems 20 times faster than a human
Ok, now make them use their own tools to build from scratch.
If their compiler is this optimized I'm sure it'll be great.

You guys who keep naysaying articles like these are missing the point.

The point isn’t whether or not these claims are valid. The point is that society does not value you and would rather continue pursuing these half-baked solutions that barely work instead of paying you.

In other words even if the tech isn’t there yet, society is showing you its true colors as to what it thinks of you and what it wants to do to you, so trying to provide value to society as a software engineer as a long term plan is something you need to start looking to exit.

>>108076936
All of those technologies increased realiability and accurance compared to what was berfore except for a small subset of problems (in which case they did not replace people), AI neither increases reliability or accuracy.

>>108076946
yup
we're holocausting their job market :)

>>108077006
this
codejeets learn their place in the world

>>108077037
cope
software companies already took a massive stockhit because SaaS is rapidly dying, you can do everything inhouse now

>>108077006
It's hard to beat the short-term value of slop, or slop for dead end applications, but in the long run, the slop will self-destruct and deteriorate everything. High level programmers are still needed for deeper and high reliability applications, though at relatively much higher price.

>>108077006
you guys worry too much about "the software industry"
it's completely irrelevant to hobby programmers what tools some retard in silicon valley decides to use

>>108077044
>that 1-2% of the market that needs autistic C/asm? yeah that will sustain our bubble of webjeetery and python
kek

>>108077065
>others suck therefore you do, too
im sorry thats your experience with programming, anon

have you tried working in construction?
theres plenty of money to be made if youre a hard worker

>>108077092
theres not much to be afraid of
so i dont count on that
youre savagely retarded though

>>108077044
Yeah but again, society is clearly signaling to you right now what it thinks of those highly capable engineers and what it wants to do with them.

@grok pull up those "looking for job vs hiring" over time graph for codejeets

File: Screenshot from 2026-02-0(...).png (33 KB, 736x167)

33 KB PNG

>>108077119
>baby please come back, i wasnt myself...
?

File: ai-lmaoo-google-commitment.png (138 KB, 867x707)

138 KB PNG

>>108077134
does that graph include india?

>>108077119
They are not signaling what they think of them, it's just that they need much fewer number of them than previously. It's not personal.

>>108077164
>b-but stock market
still richer than you
seethe harder poorfag

File: ai-lmaoo-apple.png (142 KB, 904x724)

142 KB PNG

>>108077194
suuuuure

guys, can you just not reply to the retard who is so obviously driven by inferiority complex?

>>108077044
Look at it this way, if society is clearly hell bent on getting rid of you and actively hates you, why continue the fool’s errand of trying to provide value to it that it doesn’t want?

Everyone would rather have a society where nothing works properly and everything is falling apart. I don’t know if you’ve ever had the pleasure of working with city/state government institutions, but they were already halfway to the point of barely functioning before this LLM crap even started. Disorganization all over the place, websites always down, websites always having errors when they are up, left hand never knows what the right hand is doing, nobody who works there can give you a factual answer. Now imagine how much worse it’s going to be when all customer service is replaced with chatgpt bots that keep giving you wrong info, or worse, companies plugging in chatgpt bots into core systems to make decisions, then they hallucinate that you owe a $3000 bill when you don’t and there’s no way to appeal it or get a human to look at the obvious error because all avenues of appeal are all chatgpt bots that agree with one another.

Point being that society is already apparently okay with systems that barely work and are falling apart. So why wouldn’t they mind replacing a guy who knows his stuff with a chatgpt bot that sorta works if it will save them some money?

>>108077194
I don’t know if you’re just not paying attention to what people are saying, but they all hate programmers now. Not just employers, even just normal people who work gay office email jobs are saying that they can’t wait until all the engineers get replaced by AI.

>>108077215
why?
you want him to keep polluting the thread?
at least when he gets thoroughly btfod he stays silent for an hour or two, until the btfoing gets slid from the last post
theyre not gonna stop
some of them are literally paid to shill
some are just mentally ill

if you wanna fix things rally the board to spam the irc with demands for an ai board

>>108069375
If the Epstein files happened maybe this will also happen?

>>108069225
How do we know they didn't just overfit to the training data? They never release it, and unless it does this again with unique code I do not believe them one bit.

>>108077243
That's just people on the internet that seethe about programmers for whatever reason. Some deep-rooted inferiority complex or something most likely. Actual normal people don't care.

File: file.png (116 KB, 1096x226)

116 KB PNG

>>108076712
A 1:1 translation in the style of c2rust would be full of unsafe code, this doesn't. So even if it translated an existing C compiler to Rust but made it safe that'd be a huge achievement.

>>108076757
>attention span so fried that he can't read more than one post

>>108077981
>be a huge achievement
Not really since it somehow managed to be even slower than unoptimized gcc effectively making it useless. Well I guess that's an "achievement" in a roundabout way.

>>108078450
>but it can translate gcc/clang code to rust retard
>if it translated an existing C compiler to Rust but made it safe that'd be a huge achievement.
This is a hypothetical tangent. A safe C-rust translation would be an achievement if it were that
But it isn't because a translation wouldn't have the slowdowns that we're talking about.
This goes back to anon proclaiming that it's copy-pasted without providing any source it was supposedly copy-pasted from.
You're just making up bullshit arguments to dismiss progress.

It's a crappy compiler, yeah, but it would have been impossible just a year ago. Consider how things will look in another year.
Or do you just expect things to immediately stop making progress after today?

>>108078580
Anon did not mean it "literally copy and pasted". Don't be obtuse. It took existing codebases from known C compilers and made the code rust. That's what LLMs do.

>Consider how things will look in another year.
Yeah more low quality slop like it's always been.

>>108078580
these people are going to keep making excuses
you don't have to persuade them
a few more model releases and it'll be over

>>108078644
>It took existing codebases from known C compilers and made the code rust. That's what LLMs do.
No, they have not done this two years, in fact DARPA _wished_ they could do that and started a program to reach that goal eventually. https://www.darpa.mil/research/programs/translating-all-c-to-rust

But that's not happening here, because if it were a simple translation then we'd get much better results. It could have chosen to translate tcc for example.

Here's the first meaningful commit: https://github.com/anthropics/claudes-c-compiler/commit/26f6f8b2c1db903cb718bd8d0496ccdbb9711294
Is this based on anything?

>>108078773
Yes, open source C compilers, particularly GCC as the blog post describes.

gpt beating dota 2 pros live in 2018 is more impressive than this

>>108069225
>I'm a white man, I wrote my own fucking compiler

>>108070975
>And I could just cheat like Claude did by compiling into C and then compiling using gcc.
wait what

>>108075197
KEK

>>108077246
>troll tries psy-op
so you would go away and let us have a discussion in peace.

File: 1770381674404758.jpg (7 KB, 236x252)

7 KB JPG

>>108069225
>LLMs are simply models trained on exisiting data
>is a glorified autocorrect system
>autocorrect manages to recreate an opensource tool thats existed since the late 80s and been foundational for many other technologies
>referenced in textbooks, documentation and universities for nearly 50 years
>autocorrect system is able to recreate one of the most cited and used utilities ever made
>"WOAW THIS IS HECKIN WHOLSOME REDDIT! AI IS AMAZING"

LLMs are good at explaining errors, finding syntax typos and google searches. Try doing something even slightly novel or niche with them.
Claude couldn't make SCSS fucking work with Hugo last night. Spent an hour going back and forth with the thing, then solved it myself in 5 minutes after reading the docs.

>>108074764
>Go make a C compiler on your own
pretty sure the C interpreters are like 5k lines of code.
Writing C compiler is not hard.

>>108069275
I'll ask my Claude to make the logo

>>108069307

Hello NitpickLawyer!

>>108070369 (sage)
b8 used to be believable

>>108069937
>nooo it should count only if it doesn't know shit about the subject!!!
lmao, do jet planes also "do not count" because their designers had prior experience designing propeller planes?

>>108069225
Give me $20k and 3 months and I will build x86 C compiler that performs as well as gcc. Not the crap Opus created.

File: images.jpg (6 KB, 168x300)

6 KB JPG

>>108070477
Attack on Titan's final arc. The skeleton character basically acquires a demigod's power, turns into a monster and starts a huge genocidal campaign against its enemies.

>>108069609
if it really is "over" for software engineers, they should really be developing other skillsets while they can.
although, this c compiler is total trash and not a good example of ai slop code by any means

>>108069299
Can you not read, midwit? They're testing safeguards and are pushing the envelope on purpose. R&D always takes up time + money

>>108069731
If it's JUST translating, I think it will be possible in a couple of years, maybe 5, with the size of clang codebase.
But it's already possible today with smaller programs, take this for example
https://github.com/microsoft/mimalloc/pull/1212
>translated code results in 100% test pass (test_api, test_api_fill, test_stress)
Pretty impressive.

>>108069480
A very simple C compiler does not require a team nor $20k

>>108082345
>It is worth noting that C2Rust fails to translate mimalloc because it struggles with complex atomic operations and strict thread-safety requirements. In contrast, Rustine succeeds by focusing on semantic intent rather than relying on shallow, syntax-level transpilation.
Also this one, holy shit.

>108082345
>108082459
>t. AI glazers getting paid per (You)

>>108075198
Not if a human guides it.