#programming
1 messages · Page 521 of 1
what are we talking about
The correlation between LLM autism and their efficiency to code
I lied btw 
i could guess
🪣
dear god
Have you heard of a new branch of math called gender calculus? 
have you heard of
steve jobs 
Gender studies degree
Just finished downloading 2gb of geometry dash reply files
I google translated my code
But why
Because I can't read english
But you can write
Don't write code in french
C doesn't like having é in a function name
What does feels like when refactoring 1700-1800 lines of codebase archit
If you don't have stupid teacher who want you to write code in french, no you just learn english and write english and read english
If you suck at english well that sucks
Feels like fighting final boss for me
That's smaller than my main.c
meanwhile my PR: https://github.com/weenachuangkud/FastCast2/pull/39
Yes, this is draft and it's 12247 LOC
I'm sorry bro
12k added 1k removed
we love ai slop
My final project before quitting roblox game dev
Simplest roblox script
Meanwhile the entire codebase each files was either 800-1000 lines
Does anyone feels like me
As roblox game dev, it was some furrys fault


why are we counting unsafes in a project that needs to interface a lot with webkit 
its basically just a 1:1 translation of the zig code... with all the unsafe behaviour ontop
One every 91 lines iirc
How much token does bun used for refactoring bun in rust lmao
actually i wonder now, how many does deno use
you could make a wrapper for the unsafe bits. so it doesn't contaminate the whole codebase
Also what's silly is that Anthropic buys bun a tool to convert typescript into javascript
Anthropic vibecode typescript product
They then vibecode some typescript to javascript improvement in zig
They then vibecode zig to rust
They could have vibecoded Claude code in rust in the first place
Because rusts whole gimmick is safety
2 unsafe. in the complete repo 
7... but they are basically right next to eachother

Is this new update for bun in production yet
Because i might honestly switch back to node
they basically doubled the code in the repo
its canary
every .zig has a .rs i think
not released yet
honestly if you don't need bun... just use node 
It's in main
deno
Bun was good for a while as a better node
well.. deno doesn't have full compatiblity yet....
i think?
Tbh don't use Javascript and problem solved
Tell that to browsers
No no
Javascript has a use in browser
For minimal scripts
Saving and loading permanent state to local storage
Or dynamically fetching some json to put it in a table
The only reason i write typescript/javascript is a discord bot or a website
But that's it
coderabbitai doesn't care about how bad my commit messages is
Beyond that i dont touch it
Those are valid use of javascript
I do it in Java or python for the discord bot and for the website i use plain HTML and CSS and the most minimal amount of JS possible limited to the proper use cases
Tbh i use astro for my websites currently but i dont touch react or anything
Also typescript is a sin
Where javascript shine, it's untyped
Who tought it was a good idea to have to compile my interpreted language to another interpreted language
not falling for this ragebait 
Kotlin is meh
Jsdoc is the way
OK but then write in system language
tfdym bun is a tool to convert typescript into javascript
For a website???
Is it wrong ?
very much so
No a website doesn't need code
And on the server yes use a more appropriate language
Maybe in the early 2000s
Even by now
use rust obviously
All JS is needed is to save and load data
this reads like someone who has never actually wrote anything on the frontend even once
if your full stack isn't rust what ever you even doing?
Showering
Your frontend is supposed to render not compute

contact forms:
dashboards:
literally anything that isn't a marketing page:
well.. but surely you didn't change the whole piping?
Me when HTML can do forms
Yes for a dashboard js can be used to load a json into a shadow dom template
Literally anything that isn't marketing doesn't need JS
Zig


And for Web apps use wasm or whatever
Or just don't and make a proper native app

And not some shit that require a whole Web browser and 3pcs worth of ram
As a Kotlin dev I am confused
the industry decided against that. everything is electron and you will like it 
I hate electron so much
Don't worry Java weird child
Wrong i'll fight
they can do forms in HTML about as well as webassembly does servers
I'm tired of being forced to run bs
use tauri then... which is rust 
That sucks too
It's the way
Desktop should use native UI
You're just brainwashed by the thousand framework webdev
I hate webapps
see, the thing is, if I want to try a product, I kinda don't want to have to run a potentially closed-source binary on my computer
But don't worry React guys said the DOM is slow so they actually use 2 DOMs instead
Instead of realizing wait maybe the problem is we're doing too much shit
???
And that's when we invented sandbox
i wish PWA were the thing we actually do instead of electron... would solve so many issues
and how would we get the sandbox in the first place
this tbh
That's call you spin up a VM or your favorite sandbox program
I know some workers at discord and 99% of their security issues come from electron
aaaand tell me how we get the sandbox program in the first place
pretty much everything done in electron apps can be solved with either PWAs or tauri
I wish electron wasn't as prominent
Me when Firefox added them and then removed them to instead add back glorified shortcuts later
PWAs are interesting but Chrome is the only one to do it semi-right and it still sucks
swapping electron for tauri is basically just.... one thing for the other
i don't think it solves much. its just a different way
Why take the most dated technology and stick it on the system too
Anyway one day we'll realize wait when i actually care about the assembly which is generated by all those layers of mess my shit is fast and just work
We used to have the fucking L3 cache as RAM
idr the last time I actually used a tauri app because everyone loves electron these days but from what I remember they actually worked quite well, assuming if you build them competently
And yet our shit is slower than ever
Gonna get worse with vibecoders
gotta wait for more of those DROP TABLE users happening with claude
And i didn't even started talking about the whole idea of using dependencies for everything because the web wolrd fucking love that
I have a bunch of those saved
i mean... rust too. you get like 50 dependencies minimum with any project 
do share with the class
yeah I don't get why npm is hated on for having dependencies when rust also has the issue too
python as well but I feel like most people already acknowledge python is really just a playground
One of the reason why i don't like rust
Language package managers are a bad things, it should be harder to have dependencies
And important things should be in the std
there will always be dependencies
damn I was hoping more ngl
I got more on my pc
Yes but to the extent if Web dev, python, rust, is bad
Im doing my morning routine
Honestly those 3 are the most notorious for their packages
Nuget isnt as bad
Or even gradle/maven
Imo packages in Java is roughly where it needs to be, you can do it, but it's not seamless and the std is big
hot take here: not really, maybe they really should tone it down a bit, but there will be a dependency explosion the moment a language gets thrust as the "best solution" as people build abstractions over stuff viewed as "complex", that's often why dependencies exist
but tbh why bother caring when it's funny to see another vibecoder pull in a dependency that gets compromised one week later
Npm shouldn't let u upload packages unless u have 2FA
C# imo is bit too easy
But way better than the rest
was gonna say
Alot of sidechain attacks are just dev stupidity
And the sheer number of packages increasing the likelihood
anyways, I agree with this sentiment
trusted publishing is not enough, especially with people still using pull_request_target like it's some safe way of doing privileged things on a PR
Thing is we do a lot what's easy to do
Complexity, walls, things being hard is good
It give us time to rethink reconsider
Every dependency is a cost not a blessing
Signing stuff is good too
Tbh for that there's big red text all you have to do is fucking read the doc
People using the wrong tools cause security issues absolute cinema
isn't provenance meant to (in some way) act as this?
signing stuff could mean un-reproducible builds which may be a no-no to some people
ye, I think having tooling that accommodates this is good, and I think Cargo does this pretty well
the one part of it that I don't like is that there is no real option for dynamic linking of dependencies
unlike C or C++ you can't have a separate package for your dependency and update it system-wide, you have to rely on every application to keep their stuff up to date
Like this is written in the docs
Workflows will not run on pull_request activity if the pull request has a merge conflict. The merge conflict must be resolved first. Conversely, workflows with the pull_request_target event will run even if the pull request has a merge conflict. Before using the pull_request_target trigger, you should be aware of the security risks. For more information, see pull_request_target.
Running untrusted code on the pull_request_target trigger may lead to security vulnerabilities. These vulnerabilities include cache poisoning and granting unintended access to write privileges or secrets. For more information, see Secure use reference in the GitHub Enterprise Cloud documentation, and Preventing pwn requests on the GitHub Security Lab website.
Not sure but its kind of a standard elsewhere
dependencies are trade-offs just like vendor lock-in
unless you still use undici or is-odd in 2026 then you're just stupid
Or atleast checksums
checksums work, it's also what js package managers do as well
always why I run my installs with --frozen-lockfile
Yeah they have that then idiots maintaining the git get compromised lmao
?
Referencing alot of the recent sidechain attacks
frozen lockfiles are meant to prevent that from being a big issue
when installing deps (with pnpm in my case) if you run --frozen-lockfile, any hash mismatches or updates fail the install and don't update the dependency
also pnpm finally set minimumReleaseAge to 1 day by default now which is great
Personally i run all web dev stuff on a github alt in github codespaces at worst my alt will distribute malware and github gets hacked
And i avoid the web dev
Also another problem is people trying to move too fast
Just take your time and stop doing all your slop ideas
I'm taking my time already that's why I'm stuck progressing on a discord bot
Im hoping for better native ui tools in the future
Imgui is still king
Which is kind of sad
or it just shows how good imgui is 
what about qt 
Dont even joke

Their license makes it not even worth looking at
Qt is decent I heard
but never used 
imgui definitely much simpler though
why?
Im not gonna pay for a license when i can do it for free
why would you need to pay for QT?
Real up on their licensing
qt has convoluted licensing, it's "free" but under weird conditions, and different parts of it are differently licensed, even parts that are very closely tied
i guess it depends on what you do...
i threw gpt 5.4 at the problem and it was able to create a pretty decent QT app quite easily
looks like the license boils down to "LGPL but some parts are GPL so better be GPL compliant"?
seems reasonable
i'm sure it was, but if you want to use it commercially you basically need a lawyer
Yeah its not free
forums are full of "i'm confused about qt licensing" since 2003
i only know QT really because plasma is build on it and source engine uses it
Its used for production by companies yeah
But u will almost never see personal or indie software running it
Its more for enterprise budgets
makes sense i guess
Still hoping for a better alternative one day
there's tables people keep making for it
Imgui licensing with qt quality
yeah idk this does not seem that complicated
remove the enterprise column and it's just every GPL project ever
read the note at the bottom g
even they just tell you "man just talk to a lawyer or our dedicated licensing experts idk"
if you're random open-source dev you'll just have to use the GPL licensing
if you're big enterprise you have someone who can figure out the enterprise licensing for you (or can just ask Qt I guess, you have money for support)
Imgui chads
it's also an incomplete table, depends on the distribution mechanism as well
Should be AGPL instead
agpl makes no sense, qt isn't a networked backend library
ye, AGPL wouldn't really do anything for Qt 
maybe they have a server somewhere in there but it's definitely not a focus
Boblox give me better code editor pls
What should I call my SLM 
“Moron” sounds apt but also like a self-fulfilling prophecy..
Somewhat Lame Model
I’d pick something easy to detect with STT in case you want to hook it up, maybe marry
do you need to use their code editor?
can you not use a external one?
Might be able to just open the dir in vscode or zed-editor

I’m already happily married
this is just for testing shit
What type of training sources are you using, I’ve mostly been prototyping with my own messages after de-duplicating, filtering & mixing with instruct data so it’ll listen
I have like three books worth of rpg campaign notes I was gonna use, idk if that’s enough tho :P
Maybe messages might be a better option
Jetbrains editor + lua(u) plugin or vscode if you're into that
biggest lie I seen on this channel
Like non-conversational data? Just writing about things, it will probably achieve exactly that but it won’t learn much of communication, at least if you’re using this as post-training. Might help in per-training
Yeah that will get you easy results, it’s not perfect but it’s almost. You could also try training it on fiction, whereas some models might just know how to make ads if they’ve seen the unfiltered internet
It is non-conversational, but written as fiction and does contain some dialogue
One worry is the sheer number of bizarre proper nouns, cuz it’s fantasy slop, but we’ll see :P
ive made a generic presentation a while ago on continued pretraining shit on llm which gave me good results https://docs.google.com/presentation/d/1bl9W2ZV0aQz7fswjRm8ruhMv4qgaGZJf2xj1W64q-tA/edit?usp=sharing
You could just find and skip windows that contain partial or whole words that don’t make sense
to continued pretrain from non instruct new info
That might work 
With conversational data you can just filter the responses, and it will still understand anything. But next token prediction training uses the whole sentence as learning how to respond
I'm using vscode + luau language + selene + rojo + rokit
well it simply filters the loss at the end
attention mask and shit still same
roblox studio needs to officially supports vscode lmao
Imagine have to create local place clone on local machine, what if main place updated and not local place updated, also roblox studio doesn't have something like git pull, so imagine if place was 50 kb(this is big) it would take so long to load all stuffs everytimes it updates
(Thanks to rojo, because it was made for fixing that)
But ngl roblox studio is simplest game engine
A 10 years old kid already knows how to use it
What part of LGPL do you not understand?
????
what part of "qt has at minimum 10 different licensing types and conditions depending on your usage, distribution and business size for each of their components" do you not understand?
qt is not a single thing, it's a collection of dozens of tools, frameworks and libraries
"at minimum 10"
there's 3, and usually you really only have to consider 2 
LGPL for most stuff, with some exceptions that are GPL, so the GPL obv. applies if you're using those
seems straightforward enough
Also, LGPL itself is pretty straight forward. If you do not perform any modification to QT itself , your source code can still be proprietary(which is the most use cases anyway for GUI library)
if your stuff never sees the light of day then it doesn't matter either way
if you're using qt professionally it's complicated
You just need to give a linkable file or a way for user to replace the QT that your program depends on
complete, following todo comments, this complex global test of highly interconnected application, using my complex testing framework, that uses technologies that wasnt there 10 years ago
follow style that already there, no additional tooling or syntax required, everything you need is copy-pasting with correct values
ah also you have to somehow run application in control environment and replace placeholder data with real data
make no mistakes
that what i would like to prompt to ai to finish tech debt, and just read code instead of writing it
omg i hate tech debt
test driven development on top of <complete design first, code second> not sound so bad rn
not so much of a text here
Windows Moment
we're not getting a passing grade with this one 
You need to stop doing last minute submission lmao
The trick is to submit mvp then fix stuff along the way unless you only have 1 shot, then git the mvp
I get blue screens on Artix too. Not because it's an error screen, but because my undervolt is slightly unstable and the whole system randomly crashes at times with the screen flashing a garbled random colour.
See, at least if you want to get blue screen at least make it self inflicted
Have you considered not undervolting so hard
Have you considered that it's the only way I can get the performance I need because this piece of shit refuses to clock to boost levels for more than like 20 seconds at a time
I managed to push an exta 200MHz as the new "normal" speed
And it doesn't happen too often, at least until recently
Change the thermal paste
You mean kernel panic ?
The kernel have bsod too now. And I think kernel panic would not result in that but just the caps lock (or was it num lock?) blinking
Oh, it is systemd only?
My crashes are similar to if you removed RAM while the system was on
If caps lock is blinking you have a kernel panic
What you mean is the new drm kernel panic
I swear it was? Unless I'm just completely misremembering...
Now I need to check
Apparently it's not
I think you mean this
https://www.reddit.com/r/linux/comments/1dum3f0/linuxs_drm_panic_screen_of_death_sees_patches_for/
its a new kernel feature
Where did I read that it was a systemd thing then...
Yes
why is it called like that
Maybe that was something else?
Because the kernel panics and needs to do breathing excercises
Because you can't really recover without manually rebooting
it’s a panic. in the kernel. i know. it’s shocking
like an uncaught exception
or it got overheat
I can't say I like the look of that new QR-style one, it gives less information by default
anything that results in some unexpected condition that the kernel doesn't know how to recover from other than shutting down entirely
usually shutting down is better than attempting some half-baked fix and risking the integrity of everything running on the system
Holy shit i just got the best email of the day
It is when 1 != 1 moment of the kernel

Much the same thing as a bugcheck on Windows?
A condition that’s not really recoverable from
Wait I forgot. Does GPL allow the source code to be redistributable?
Calling an ai(s) MDASH — 

it doesn't impose any restrictions on the source code IIRC
you just have to provide the source code along with any binaries you distribute
Yes but, can the person that receive the source code actually redistribute it?
And also for that matter, can the binary be redistributed by the person?
Same 
Because damn that would be a massive loop hole
for source code yes definitely
You may copy and distribute verbatim copies of the Program's source code as you receive it, in any medium
literally the first clause(/paragraph? forgor) 
🦊
Ahh good to know 
Each time you redistribute the Program (or any work based on the Program), the recipient automatically receives a license from the original licensor to copy, distribute or modify the Program subject to these terms and conditions
this includes binaries probably
software licensing 
Such is the world we live in
pog
i just want some vacation for my brain after this exam ngl
Does anyone know how many polygons per second can an Nvidia H200 theoretically process?
Considering T&L only.
Nvidia no longer publishes polygon per second ratings for their cards.
Contrary to popular believe, this is terrifying because lipo burn spectacularly
rasterization in big 2026 smh my head, AI TOPS matter more
why are we trying to process polygons on a card that has no support for graphics apis and no display output 
but either ways, number of polygons per second does not matter nowadays
it's probably like 10+ billion polygons per second
gpus are just too powerful to be evaluated in such metric
mfw a 4080 ti has higher theoretical fp32 throughput than an h200
I'm just trying to figure out how many polygons could you theoretically render using a sentient robot's brain.
For a story I'm thinking about writing.
they would probably be using ray tracing instead of rasterization 

H200 doesn't really do polygons
fun fact, a gpu can do over 100 billion bloom filter lookups per second
You can't exactly determine a hard "polygons per second" number any more because it depends so much on a billion factors
It depends what the exact polygons even are
You see, the robot's head would be landing at a 90s game programmer's desk.
don't behead the poor robot 
H100 in the 90s is crazy
it does have ROPs though so I guess for a quick real-looking estimation you could take some card that you do know the polygons per second for and just linearly scale to adjust for the difference in ROPs and clock speed
surely accurate 
they would probably try to fit some vacuum diodes inside
hopper is basically ada lovelace++ 
*H200. And I was going to multiply it by at least 4 to account for the processing power needed for AGI.
anyone know a place where one can profile a b(2|3)00 
aws 
at that pt I'll just make up a reasonable number
i think aws only has instances with 8 gpus 
let's say 76.3 billion polygons per second 
i'm not paying $100 an hour or however much they want
I guess for the 90s that might as well be "unlimited".
Game systems back then didn't go past 300,000 polys/sec.
well, that depends entirely on how you procces them. are you texturing them? are you using an index buffer? are you doing instanced drawcalls?
in reality, you should not use a H200 for this
this was meant to be a joke... the b200 instance is $114 an hour apparently
they have a single GPU instance for the H100
$33 for a spot instance tho 
i don't need an h100 i have access to one already 
wanna find out why my code uses 255 registers on the b200
not for H200 or above I think 
I guess you could always use one of the cheap GPU clouds like Runpod
but their performance is a bit inconsistent

Geometry Engines: 0 functional raster engines
Triangles per Cycle: 0 triangles per clock
Polygons/Second: ~0 (Requires software emulation)
Real-world Graphic: Fails to match an AMD 680M iGPU

ye urcooked 
expect to pay big for bare-metal instance
with 8 GPUs because who in their right mind would put less in a node
Rent NVIDIA GPUs by the minute with live pricing, bare-metal access, SSH root, and a dedicated IP
B200 192 GB Blackwell Large-model training, FP4/FP8 inference $2.25/hr
wait, this might be huge
I mean there are people in China who used H100 to play pubg and got less than 10fps 
https://www.spheron.network/gpu-rental i've never heard of this website before tho
every other b200 i've seen was at least $5 an hour
Do you guys have recommendations for how I can optimize my GPU use? Zero by herself is 7gb flat, with VTS running, 10.5. This leaves very little headroom for any kind of gaming I want her to do since my GPU is 12, and I STILL need a model for tool use. (Vision runs on CPU)
what gpu?
4070ti
This is not including stuff like OBS or target games working. Realistically if I can get her walking around in minecraft that alone would be a huge win
apparently they did blockchain stuff before? and now providing GPU compute?
do they just jump on whatever is trendy or what
seems like you might want a second pc for capturing
I have a laptop I can probably use, but I'm not sure how it would tie into the loop
It's solid enough for this hypothetical application though
or just, run a heavy quantized model of whatever you are using
So that's how they make caching so fast these days
Atp sacrificing brainpower isn't doable, she's in a sweet spot.
ah yes, lemme just grab an R100 real quick 
keep in mind you will be absolutely crippled by the pcie bus
Didn't know deepseek made a GPU
this is assuming the data is already on the gpu
if we can just fit everything into vram... 
I'm genuinely thinking about getting a 24gb card atp
It would solve alot of problems
okay them offering spot instances and dedicated ones makes me think that the ssh root access is only for the reserved ones, since the dedicated one still says it's a virtual machine 
You know, it most likely is if you're on the scale of google
dual 3080 w/ nvlink setup
just put a whole SoC on the GPU die
share cache with GPU 

What model base
But yeah, the query is limited by the PCIe
Could I do this with 4070tis?
Llama 3.1 8b
But model is negotiable. It's set up to be hot swappable
there are no nvlink sockets for consumer use after 40 series I think
if you're google you'd do this all through ASICs anyway, whether that's inside a network nic or whatever else you have
Yeah
Well, no
The base model is 7gb alone
Q6
8b
save the smol filter inside a dpu or sth and stream queries through it however you want 
I could maybe move to Q5
Just use q4 but newer llm
Suggestions?
Gemma 4 4b
Which is actually 8b
And has mtp so gonna run faster
Which is important for live apps
now we have quantum computas, would they be able to do it faster 
yea that's fair
Nvidia does list the H200's pixel fillrate (it's 42.84 GPixel/s).
I'm lowkey worried bc it's a reasoning model and the latency on those is abysmal
I have the page pulled up though
they are itself probabilistic machines so i was wondering
it looks like there's a problem with relative paths in my libraries
there's include thingies that are using absolute file paths, and it's breaking them all

for instance, there's files like websocket/config/example.hpp trying to reference websocket/common/thingy.hpp
but they're using the full file path and clang is yelling at me for it
fix your include paths to tell clang to look in whatever folder websocket is
0_0
how
you can just perform a naming space refactor on the included file paths
no need
just specify path of include directory like Toast said
the main issue is the H200 lacks DX12 and vulkan optimizations in their drivers to limit latency and such. H200 only has 24 ROP, there are 528 TMUs (more than a 4090 but clocked lower) probably for data parsing, but these lack L1 texture cache and consumer format decompressors. also there arent even any display outputs on a H200 so you'd have to feed the render output back over PCIe to another card, and a H200 would suffer even more than normal GPUs from severe quad overdraw. you can't run stuff async since H200 expects big batch queues. You'd also suffer form WARP divergence more. And the main thing making the H200 such a valuable card, the tensor cores, will be barely used if at all.
There's also no ray-tracing cores so BVH traverse would need to be handled in software too.
the reason the pixel filrate is 42 GPixels is just 24 rops x 1785 MHz
to put that in perspective, the 4090 has 443GPixels
I don't work with clang so I didn't know
I just do it the slow way - pressing some refactor buttons
if there's an include folder generally you want add that until everything is resolved
it's not a clang-specific thing, the include path is configurable on pretty much every C compiler 
this is just how C programs in general are usually built
visual studio does all the handling for me 
ideally everything within the library is self-contained and references each other properly unless it's just broken
js install the C/C++ extension 
beri easy

just don't be that person that puts everything in a single namespace across multiple files so that it gets resolved by the compiler but any language server is having a heart attack
tho another C compiler is a lot harder to config 
we need to ban IDEs so that this doesn't happen
what the heck is an include path

You can't NVLink cards after 30 series
every time there's an #include <some/file> the compiler checks through all include paths and tries to resolve it from there
Dang. Well I'm picking up a few variants of Gemma to test out.
When you do #include, where is the file actually located?
That's include path
this is one way of adding a path to the list of paths to check (yes, there are way too many ways of doing the same thing)
wait I think I was just cloning instead of actually getting the releases
r u testing via ollama?
notkonii are you konii
or llama.cpp
I like the simplicity and the fact that in C #include is just copy-paste the file content
or-
oh dear lord I do not know what I'm doing
Yeah LM studio, but even if I wasnt I have to edit my backend to test new models
mhmm
idk if i'd use the word "like" personally
should've used love 
so basically your usual basic C project structure is kinda like this
src/myfile.c
include/mylib/myfile.h
Makefile
you'd add the include/ directory to the include path, so that #include <mylib/myfile.h> searches in there and finds the header
having the mylib subdirectory avoids conflicts with any other libraries, since those will all share the same include path
So I see Gemma 4 has vision, is that comparable to the vision capabilities of Smol500M or is it going to be more taxing?
The actual VRAM size is about the same.
how do I get a thing off of github without having to install every dependency separately
depends on the thing
what thing
releases if you're lucky and it exists usually
Usually they involve a dependencies TXT you can batch install
big if
Please understand that I've been dealing with Nuxt hydration issue for a while which makes C simplicity seems wonderful
Hehe, pp
eye rolls
looks like the only thing to install is boost
is it bad that pp just makes me think of powerpoint 😭
gamedev-slop
README explains it
nlohmann json is something you'll just have to get from elsewhere, like from your package manager or from the GitHub releases
websocketpp is included as a submodule so it's already there if you just clone with submodules
idk where you get Boost 
system package manager i guess
nlohmann is a library now? I thought it was header only?
it is
it is
it's just also not included
How times changed
you download it and put it in include directories

how do I clone with submodules
it hasn't been doing that for me
if you already have the project cloned then what you want is git submodule update --init --recursive
git clone --recurse-submodules
include directories 
-# this project structure sucks
dw we've all been there
Don't worry, at least you're not dealing with web frontend
i still forget about cloning with submodules and then my first thought is to delete and clone again 
Hydration error 
I literally have no idea what that is
you can just directly -I the Boost library's include dir. I just put it in Windows Environment Variables directly, and let visual studio do its job... again
well ye
you just need to obtain Boost somehow
and idk what the best way to do that is for whatever platform they're on
boost is not header-only so it's annoying
Me to Nuxt right now
I did download boost, not sure how to install it though
oh, and in this case you can't just put it on the include path
it CMake project
-I $(nix build nixpkgs#boost.dev --print-out-paths --no-link)/include -L $(nix build nixpkgs#boost.dev --print-out-paths --no-link)/lib 
I still can't understand why CMake didn't just make the CMakeLists.txt be an instruction to be included easily
it's still possible to switch to makefiles from this early stage, surely 
it is always possibel to make makefiles 
I hate cmake so much I just make my own batch files to compile via MSVC Build Tools
i love the github ui 
is there a way to clone the sdk from github with the dependencies
yeh
unfortunately no, unless someone fixes it
though
should be easy enough to fix i can do this real quick
this is a large part of package managers exist
when you install something though a package manager they go and get all the dependencies too 
there's no way this SDK is packaged though 
no, PowerShell is fine, that's different 
on Windows you'd use NuGet for fetching C++ dependencies I think?
Huh, I thought it was for application only
there's applications on NuGet? 
or are you confusing it with WinGet
Ahh WinGet
?
You can install CMake projects if the library author didn't do anything too weird.
find_package won't find it though
U can turn off reasoning
✅
winget can also fetch msvc++ redistribute packages 
Is there a way to hybridize it? I'm seeing it has tool call which my DarkIdol llama didn't, so it would be handy for it to have reasoning there.
technically not applications
But for personality responses, maybe not so much
I still wouldn't call that library for development purpose tho
It's like .so without the .h
I'm gonna do some testing on Gemma 4, I appreciate the heads up to look into it. If I can get it to work the way I need I might be in a good spot tbh.
It saves me the need for an extra model for tool call while she's gaming/controlling OBS
you can fetch Windows Driver Kit/Windows SDK
does that count
Though admittedly I'm a tad worried about personality degredation, I still got a DarkIdol variant so it should be roughly the same

Sure, yeah I guess that one counts
Have the LLM call a tool that turns on thinking
Or turns off
Okay fire I can easily wire that up
Relatable
@stark needlebruh your name on here is koniipilled
🇩🇪 🇨🇳 ✅
"koniipilled" 
koniipilled
knrioi
🪹
:omemga
:omemga
✅

I define koniipilation as non-English characters in names

and doing reacts
of course
Am I chaypilled then?
the only chaypilled one is konii
𓌳𓄿𓂝 𓊪𓅱 𓅨𓂋𓏏 𓇋 𓋴𓄿𓄿 𓅨𓂋🪣 🪹 ✅

Why does my phone have egyptian hieroglyph installed
is that how programmers communicate?
Yes, specifically APL programmer
That, or BQN
@rancid ore idk if you still need it but i fixed the thingy (cmake now downloads and builds boost), though idk if it works under windows (i think it should?)
https://github.com/tdortman/neuro-sdk-websocketpp

funny enough they need an older version of boost cause some asio stuff got removed 
In favor of coroutine?
Olympiadane is a mechanically interlocked molecule composed of five interlocking macrocycles that resembles the Olympic rings. The molecule is a linear pentacatenane or a [5]catenane. It was synthesized and named by Fraser Stoddart and coworkers in 1994. The molecule was designed without any practical use in mind, although other catenanes may h...

thanks!

Oh, uh, I don't know if I have Cmake.
Can't even understand what does that even mean
./build/NeuroRPS
[2026-05-15 20:26:42] [info] asio async_connect error: system:111 (Connection refused)
[2026-05-15 20:26:42] [info] Error getting remote endpoint: system:107 (Transport endpoint is not connected)
[2026-05-15 20:26:42] [fail] WebSocket Connection Unknown - "" / 0 websocketpp.transport:2 Underlying Transport Error
Connection failed!
[2026-05-15 20:26:42] [error] handle_connect error: Underlying Transport Error
[2026-05-15 20:26:45] [info] asio async_connect error: system:111 (Connection refused)
[2026-05-15 20:26:45] [info] Error getting remote endpoint: system:107 (Transport endpoint is not connected)
[2026-05-15 20:26:45] [fail] WebSocket Connection Unknown - "" / 0 websocketpp.transport:2 Underlying Transport Error
Connection failed!
idk if this is bad but i'll assume this is not my fault 
oh it expects a websocket server on port 8000
Something about a molecule with interlocking loops that aren’t actually bonded to each other which is really cursed
guys am i crazy for using this deno permission set ?
no
What do you guys think about custom automatic gc or manual gc in high level languages like lua, C# (game dev topic)
as a heads-up, in order to build it you want to run
cmake -S . -B build
and then
cmake --build build
and then run ./build/NeuroRPS.exe i guess 
Does gc reduces ram usages right?
(I need to read more docs lmao)
GC does does Garbage Collection
AKA freeing RAM which should be freed
so that it can be reused for other stuff
-- Building for: NMake Makefiles
CMake Deprecation Warning at CMakeLists.txt:2 (cmake_minimum_required):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.
Update the VERSION argument <min> value. Or, use the <min>...<max> syntax
to tell CMake that the project requires at least <min> but has been updated
to work with policies introduced by <max> or earlier.
CMake Error at CMakeLists.txt:3 (project):
Running
'nmake' '-?'
failed with:
no such file or directory
CMake Error: CMAKE_C_COMPILER not set, after EnableLanguage
CMake Error: CMAKE_CXX_COMPILER not set, after EnableLanguage
-- Configuring incomplete, errors occurred!```
holy wall of text
why is nmake there 
idk
oh it's using nmake as the default target
I still don't have the websocket thing, is there something else I need to do after cloning it?
git submodule update --init --recursive (i made it fail the build with a message to run that if it's missing anyway)
if you run cmake --help there should be a list of generators at the bottom, what you probably want is whichever one mentions visual studio
and then run cmake -G <whichever one it is> -S . -B build (delete build first)
Submodule 'websocketpp' (https://github.com/zaphoyd/websocketpp) registered for path 'websocketpp'
Cloning into 'C:/Users/vest/GoofyGames/votvMod/neuro-sdk-websocketpp/websocketpp'...
Submodule path 'websocketpp': checked out '4dfe1be74e684acca19ac1cf96cce0df9eac2a2d'```
that look good?

🪣
time to install ninja to make life easier if you already have clang
how do I do that
do you have winget
yes
winget install -e --id Ninja-build.Ninja
got it
idk where it dumps its files though, but you do need access to the ninja.exe in your terminal 
i think it should be somewhere under C:\Program Files\Ninja\ 

if you're on powershell it might already be there
where should I run that
terminal
oh yeah, I got something
CommandType Name Version Source
----------- ---- ------- ------
Application ninja.exe 0.0.0.0 C:\Users\vest\AppData\Local\Microsoft...```
I ❤️ xmake
nice it did its thing
oh ok
now run git pull in the project, i made some changes 
I like nob 
i wanna say cmake --workflow --preset clang should handle everything to setup and build it then but who knows with windows
See, the creator is perfectly sane person 
windows in the big 26 😔
CMake Error: Error: generator : Ninja
Does not match the generator used previously: NMake Makefiles
Either remove the CMakeCache.txt file and CMakeFiles directory or choose a different binary directory.```
AAAAAAAAAA
yeah delete the build folder
oh
and try again
You need to delete the build folder first
Why cmake doesn't do that on its own is beyond me
i hate this shit already
vite is asking for process.env
so i can't just allow whitelist
i'm forced to allow all
CMake Deprecation Warning at CMakeLists.txt:2 (cmake_minimum_required):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.
Update the VERSION argument <min> value. Or, use the <min>...<max> syntax
to tell CMake that the project requires at least <min> but has been updated
to work with policies introduced by <max> or earlier.
-- The C compiler identification is unknown
-- The CXX compiler identification is unknown
CMake Error at CMakeLists.txt:3 (project):
The CMAKE_C_COMPILER:
clang
is not a full path and was not found in the PATH. Perhaps the extension is
missing?
Tell CMake where to find the compiler by setting either the environment
variable "CC" or the CMake cache entry CMAKE_C_COMPILER to the full path to
the compiler, or to the compiler name if it is in the PATH.
CMake Error at CMakeLists.txt:3 (project):
The CMAKE_CXX_COMPILER:
clang++
is not a full path and was not found in the PATH. Perhaps the extension is
missing?
Tell CMake where to find the compiler by setting either the environment
variable "CXX" or the CMake cache entry CMAKE_CXX_COMPILER to the full path
to the compiler, or to the compiler name if it is in the PATH.
-- Configuring incomplete, errors occurred!```
i love windows native developement
so i'll need to manually deny
I don't know!
time to run winget install -e --id LLVM.LLVM i guess 
Y know, maybe it's a good time to learn about build system beyond IDE 
maybe 
WSL 
i think there is a computer in your error
it installed correctly but the cmake --workflow --preset clang thing still failed
The issue is not with build now
It is clang
seeing all the Windows issues again makes me very glad to be on Linux 
time for a pizza break 
thank you #baking for reminding me
so what are you even trying to do vest
I need some way to implement the neuro sdk in unreal blueprints
and in order to do that, I need the c++ version
Hmmm, NeuroSDK technically just a spec
So you can use any websocket libraries and json to achieve that
I found this header only stuff
https://github.com/zaphoyd/websocketpp
C++ websocket client/server library. Contribute to zaphoyd/websocketpp development by creating an account on GitHub.
but if it doesn't already do websocket then it's going to be easier to just bind to the neuro sdk probably
And nlohmann is header only too
there's plugins, but I'd prefer to validate before going into blueprints, because LATENCY
otherwise you still have to write bindings but now you also need to reimplement the Neuro SDK for no reason
why were you doing this anyway
because every other integration has been for games that don't run on unreal
and blueprints will save slightly more of my sanity than trying to actually learn c++
gotta love latent spaces
well.. but the cyberpunk integration is c++ so.. that would be a good starting point to see how that would work
you could also look there to see what library are used for websockets
so you're doing this for fun?
truee
i tried to learn c++ and my brain got fried and gave up
looks like, as expected, you need to write C++ to create bindings for blueprints
so you're not really going to get around learning that 
theoretically
yeah but I might be able to just kitbash that
all I really need to do is logic

this uses a C Neuro SDK, not the C++ one
build process is even more cursed than the C++ lib though
but mabe easier? idk

would it still work with unreal?
someone rewrite this library in zig. so i can then rewrite it in rust
probably 
as long as there are no conflicting symbols 
This is fork of original library
As if gamedev is not cursed enough already
Benefits: actually works with Neuro backend
I mean, just look at Sam
Ask Claude
pretty sure she made that day 1 lol
if someone else wants to set up a sdk for unreal please do, I don't know what I'm doing
NotNeuroSDK? 
Ohh, surprisingly normal name

rust moment
oh no
I take that back 
oh no, what the heck
i don't think there is a zig library yet
oh yeah you probably just can't do this with the type system at all
macro reasonable
So, wanna follow the pyramid? 

what the fuck
idk what GAT is but variadic tuples do not exist in rust
i would... but idk if unreal even runs on linux and i also don't want to download 100gb 
ye, I confused it with something else 
What the hell is that
it's time
C++ is calling you
Claude oneshots
As tsoding said, you need a muncher for that 
hi superbox
✅
https://dev.epicgames.com/documentation/unreal-engine/linux-quick-start?application_version=4.27
there's this, but the whole download thing is understandable

hi Sam
Damn, I thought you are still away 

I've heard making mods for votv is a pain on linux though
compiling the engine..... i hate waiting 3 hours before actually being able to do anything...
WHAT?!
I wouldn’t wanna be the guy debugging video game mod with winedbg
Did they actually require that to do unreal dev on linux?!
UE5 took like 3-5hours on my pc to compile 

looks like it 
ono
That is beyond cursed

You weren't kidding 
Currently, we don't supply a binary installer of UE4 for Linux users. The good news is that you can compile a binary of Unreal Engine from the same source code that we use to develop the engine.

i think it works
(very fragile though would not recommend)
Maybe is of the libc issue?
GOOD NEWS???
As they say, Win32 is the only stable Linux ABI

there is also
syscalls
Quite
Huh? I thought linux ABI is always stable
the syscalls are
Sure userspace ABI may be unstable but the syscall are stable right?
nothing else is
Ye
doesn't matter if everything goes through glibc
Embed the libc as static linked library 
just rewrite libc in rust. that will solve all our problems 🙏
Don’t you still have GUI stuff if you wanna live outside of terminal
Embed the libc into everything duh 
Become rust
Statically link everything
Rust slander
this is Go tiers of static linking
AINTNEURWAY
wait i just found out there is a rust screeps starter template... https://github.com/rustyscreeps/screeps-starter-rust
Who was it that wanted to play screeps using rust but ended up needing to pull some DOM library 
-# also it would be nice if we didn't have to statically link everything and could just figure out proper package management and build tooling but that's utopian
Honestly nix got it right IMHO
That, and bsd firejail
I think there's still a lot of improvements possible but yeah, it's better than a lot of other stuff
sadly not everyone agrees 
Everyone just uses the ancient mp3 format and ancient gif format and ancient h.264 format
They love high file sizes and shitty quality for no reason
i'll be honest. i don't think its worth using a wasm based screeps code base.. unless it can compile to JS i don't think its worth the overhead
okay so @rancid ore
winget install -e --id MartinStorsjo.LLVM-MinGW.UCRTto install the new version of clang- then restart terminal to make sure the added aliases are usable
- run
git pullonce again - make sure to delete the
buildfolder cmake --workflow --preset clang-mingwshould hopefully work
…can’t you just install Visual Studio?
I see
(you should probably just use visual studio tbf)
ah yes.. downloading a 10gb program for a 1gb library 
Naah, f that
Or just the build tools, CMake should handle it fine anyway
Toast is using cmake, yes
installing VS build tools would be reasonable, ye
vscode sure, but never VS, never again 
I hate luau, I hate coding, I hate ml, I hate embeddings, I hate file io, I wanna move into the wilderness
is my deno.json sane ?
Unless you are makin .net gui apps
just dont
jokes on you it's only 700M 
Install Zig and use zig cc 
unironically not wrong
They do make it easy to develop GUI program tho
if only you were this based when it comes to indentation styles

look
im usign the incorrect way for my nix cofig for parity
you have to give me some credit
Konii level of changing space to tabs
I'm gonna continue hating embeddings now
OK nevermind I still hate this
i wonder if there are new devs that are out there still learning c
definitely
Not willingly
Electrical Engineers still uses C a lot
whenever i think of learning c.. i think why c when i could learn zig! but zig is unstable... so c++.. but why c++ when i could use rust!
and then i end up using c#
Since embedded system tooling are VERY slow to catch up
C is just too useful to not know at least a little bit
if you're doing any systems programming at all
if web then whatever
The joys of doing both
Just give it 62 years and then they will finally start using lua5.1
i tried making my own voxel engine in c++
it used 16gb of ram and had a memory leak
Use Chrome btw, since firefox doesn't have webserial api
ok have you considered
no
But I can use my phone to look at modbus data without needing dedicated apk 
because HOLY HECK that's a massive privacy risk
it doesn't work
Plug in OTG and USB-RS485



You stole my 3%












