Thread 106214357

Why does Rust buckbreak AI so hard?

>>106214357 (OP)
there's little Rust code for LLMs to be trained on
same goes for COBOL 2 and older Fortran dialects

>>106214357 (OP)
>I am unable to complete your request
What a mouthy little cunt. I wish he had a physical body I would beat his ass.

lol failing on an import

>>106214357 (OP)
if you want to shit out code, use a language appropriate for code shitting.
and if you can't tell what's wrong, not even real intelligence can help with this level of retardation anyway.

>>106214750
>shit out code
>not even real intelligence
rust was specifically made for people like you to shit out code that they don't have the real intelligence to be trusted with languages like C.

>>106214357 (OP)
no one understands it so it can't learn from those that do

>>106214477
A hundred billion more to Altman and he will fix it, I promise!

>>106214357 (OP)
Because for AI to master ugly, complex, and unpopular languages, it needs human-level or AGI-level deductive reasoning, sirr.
Expert models don't have it and they require a lot of good data and adjustments, which Rust does not have.

>>106214896
>more sophisticated machines are for stupid workers
your min wagery smells

>>106215883
they are though, most skilled and highest paid labor relies more heavily on human factor, rather than machines

>>106214357 (OP)
Glicko-2 is a meme, use the full-pair Brandley-Terry model from a paper called "A Bayesian Approximation Method for Online Ranking". It natively supports 1v1s, team-vs-team, free-for-all, like TrueSkill, with predictive performance similar to TrueSkill, but without the patents that come with TrueSkill

>>106216337
>is a meme
>performance similar to TrueSkill, but without the patents that come with TrueSkill

>>106216353
What's the issue, Glicko-2 works fine for 1v1 but heuristics that people use to make it work for team-vs-team games suck. Both TrueSkill and this method support team-vs-team directly, and it's really better than Glicko-2 heuristics, but team-vs-team player ratings and matchmaking in general always sucks either way, it's too difficult

>>106215970
>manual labor is peak skill
the min wagery continues to smell

>>106216610
Where did I say anything about it being
>manual
?

>>106216337
>use thing from the 1920s
No

>>106214896
>rust was specifically made for people like you to shit out code
Quite the opposite. Rust is very rigid and formal, you can't just bodge your way through it.
Python is a language made to "shit out code".

>>106214357 (OP)
Can't blame it when it comes to Crust

>>106214357 (OP)
I've noticed that AI's seem to like giving unsafe Rust code unless you preface your prompts by telling it explicitly that you don't want unsafe code. Otherwise it will sometimes give you unsafe code, tell you it's unsafe, but give it to you anyway.

I'm guessing as other anons said, there is really not a lot of rust code for AI's to learn from compared to other older languages - and half of what it does find is probably C/C++ devs doing it "their way" with what rust considers unsafe

>>106216696
>can't follow a couple of arguments to their logical conclusion
i apologize to min wagers

>>106216765
Rust is midwit "if it compiles it's bug free!" mentality.
>>106216959
It gives unsafe code because doing actual work in Rust requires unsafe.
So the whole training set of actual code is unsafe.

>>106217189
>I can't write safe code
Got it

>>106217189
>It gives unsafe code because doing actual work in Rust requires unsafe.
Give me an example of what unsafe code can do that safe code can't

>>106217189
>Rust is midwit "if it compiles it's bug free!" mentality.
How is this related to your original claim?

>>106217126
nta but you're acting like a fucking nigger. is min wager the only comeback you know?

>>106216337
why doesnt valve use it then if it's so good

It's also dogshit at helm charts

>>106218391
He's right though. Minwage cucks act and behave like he describes. Truth hurts.

>>106217747
>Give me an example of what unsafe code can do that safe code can't
Make a system call.

>>106214357 (OP)
chatgpt managed to get me a good bit into my project but had to constantly change a lot of shit for any extra feature. I expected Rust to be more condensed code but shit needs a lot of code

>>106218952
i guess it's slightly easier when you can at least tell if the llm hallucinated a crate or a crate API. that's the lowest bar imaginable that op somehow still managed to not clear.
the code will still be pure crap, mind. i remember a noob in another platform sharing his happiness because an llm was helping him "learn rust". so i asked him to explain a snippet from the "working code" he happily shared, and he was like "yeah, i understand". he didn't of course. because he thought the code was doing something wrong. it didn't. it was just gibberish verbosity that did nothing and didn't need to exist at all, which goes to your second point.

>>106218852
Neither can virtually any high level language, without use of non standard extensions, intrinsics or straight up calling low level code. Even C doesn't have any standard way of invoking a system call.

10 years, 1 trillion dollars, and GPT12 will finally be able to parse Rust code

>>106219518
Irrelevant to the question.

>>106218952
ChatGPT feels like it is generating C/Java code and then transpiling it to Rust. The code it outputs is not idiomatic and has tons of pointless shit.
I reduced a function my friend generated from 30 to maybe 10 lines of code simply by using iterators and option/result correctly instead of writing it all out imperatively.

>>106214357 (OP)
>I am still unable to resolve this compilation error
Shoulda put a threat in your system prompt, he would try harder.

>>106214357 (OP)
because it's overly complicated language that only mentally ill people can understand.

just wrangling with the borrow checker requires great mental energy, working memory and reasoning skills.

>>106216738
It's like you didn't even bother to understand anything, "A Bayesian Approximation Method for Online Ranking" was published like 10-15 years ago and it derives three models, a Brandley-Terry one, a Plackett-Luce one and a Thurstone-Mosteller one - the Brandley-Terry one turns out to be the best in benchmarks. Elo and Glicko themselves are related to the Brandley-Terry model. This paper is also what the OpenSkill library is based on.

>>106214357 (OP)
In theory, you would never need to emit rust, AI should be able to emit C code without the issues Rust purports to solve.

>>106220520
>oh I got another segmentation fault let me burn credits looking at the core dump

>>106219564
> feels like it is generating C/Java code and then transpiling it to Rust
That’s because 99.999% of *all* rust code was generated that way.
AI is a competitor to Rust — both are different ways to launder IP.
“I wrote it in rust” is practically announcing to the world that you stole the code and deleted the attributions, and changed the license so it can be re-stolen and made proprietary by other companies.
Time is of the essence before the courts catch on to the Rust scam.

>>106220806
>announcing to the world that you stole the code
I've got news for you that you might want to sit down for

>>106218420
I assume they don't actually use Glicko-2 as-is with some heuristic to make it work for team games, but rather a modification of Glicko-2 for team games that makes sense mathematically. But I don't really know exactly, didn't Dota 2 literally use Elo (or a modification of it, presumably without any kind of deviation or uncertainty values, like in the original) until very recently?

>>106216409
Open/TrueSkill suck balls when teams have roles (TF/ET-like) unless the player picks always the same role. You have to add heuristics that destroy the point of memeskill

>>106222219
But isn't that an issue with every rating system

>>106223125
But if you're gonna tune it for that, better use the best building block for it

>>106220806
Nothing you have said makes any sense.

>>106214357 (OP)
It gets stumped by fairly simple Haskell bugs too. Goes to show that these models have not advanced on a technical level after several years and billions of dollars. Despite the companies using words like "thinking" and "agents" these are still stochastic parrots who can't generalize for shit.

>>106214357 (OP)
wtf I love rust now