[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
▼ Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: IMG_8029.jpg (53 KB, 730x487)
53 KB JPG
This is it. This is the holy trinity of vibeslopping.
You can give the worst agents vague instructions and as long as you tell it to use elixir first, go second and rust when performance truly demands it, everything will come out perfectly.
>>
>>110283957
These are 3 VERY different language ecosystems you've linked here. One of these is highly oriented around web dev, another at app dev, and I don't know enough about Erland to know where it's carved out a niche.
>>
>>110283957
forgot typescript

LLMs love typescript
>>
>>110284035
The “TS and Python are the best languages for LLMs because of the training data” takes are misguided at best. Ecosystem also doesn’t matter anymore
>>
>>110283957
The best I can do is writing descriptive paragraphs about what I want and having Opus multipart them in pristine bug free C++ targeting g++ with a tiny efficient Makefile
>>
>>110284053
What would you say are the best languages for LLMs then?
>>
>>110284080
The three in OP. Gleam will probably overtake Elixir at some point for the BEAM language of choice, but not there yet.
>>
>>110284035
>>110284053
Must be types then, JS doesn't have types kek
>>
>>110284105
Elixir barely has types but it still works insanely well for vibeslopping due to immutability
>>
>>110284090
I've never even heard of Elixir and Gleam. I've been trying to vibecode something complex and I've tried C, Zig, Nim, and Rust (in that order) and all of them have only been partial successes, but end up getting worse with time rather than improving, hence why I drop them and start from scratch. I'm skeptical that switching to a different language again would help, but at the same time I'm intrigued.
>>
>>110284144
Neither is a fit for low level stuff so they probably won’t work for whatever you’re doing, unfortunately. They are god-tier for anything high level though.
>>
>>110284144
>end up getting worse with time rather than improving
That's not the fault of the programming language, anon.
>>
>>110284080
>What would you say are the best languages for LLMs then?
Ecosystem doesn't matter any longer for backend shit, just tell it to use Rust for everything because there is no point in using anything that isn't efficient low level languages, the only reason C doesn't win out is because Rust allows crate imports which minimizes token usage. If it weren't going to be a horrendous waste of tokens you could be telling it to use assembly, it literally doesn't matter any longer. Why take the performance haircut for python at all? Python was for getting stuff out quickly and easily where performance wasn't too much of a concern, now you just tell the AI to build the shit.

AI changed a programmers job from hunting and pecking through bizarre syntax to focusing on architecture and problem solving for the most efficient path. If you give an AI a shitty solution to a problem it will make a shitty program, describe a properly designed program with an emphasis on efficient solutions and you get a good one.
>>
>>110284178
Yeah, I just looked up Elixir and apparently it needs a VM to work. That sounds absolutely dreadful and pointless.
>>110284188
I know that. It's the fault of the AIs...
>>110284204
That's pretty much the conclusion I came to and that's why all the languages that I've been trying have been low level. The problem is I'm not a programmer myself, so I can't always find the proper solutions to instruct the AIs. And believe me, Deepseek needs a lot of instructions.
>>
>>110284243
>And believe me, Deepseek needs a lot of instructions.
Just use Claude for this generation of coding models, maybe next gen the pecking order will change but atm Opus 5.5 on Medium for complex designing and Sonnet 5.5 on High for basic bug squashing. Using a shitty model and harness is what digs you into problems you spend time and tokens digging yourself out of.
>>
Name one (1) application that did this and is successful.
Successful means actively used to this day not pulled 10k likes on xitter and vanished one day later.
>>
>>110284144
You know you can tell the AI to refactor the codebase and look for problems, right?
>>
>>110283988
Concurrency
>>
>>110284366
Using Claude might work but I'm too stingy to pay for AI.
>>110284422
It's not a question of "problems" in the codebase. Usually when I have an issue, the code itself works, but it just doesn't perform to the standards that I need.
>>
>>110284562
>Using Claude might work but I'm too stingy to pay for AI.
I was the same way then when I got a free month for OpenAI this summer it was like getting shackles struck off and I started using it nonstop to unfuck FOSS programs and just do stuff nonstop even outside of the coding stuff.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.