programming

[olive sable]

actually, im pretty sure those arent commercially for sale

i know 16K DLPs exist, but thats research grade

[split steppe]

i just want 3x 1440p 16:10 30" LCD monitors for productivity, and then my 4k 42" oled i think is perfect for games already

[olive sable]

the commercial ones use 4 8K projectors to do 16K via EPPT

[olive sable]

seems reasonable ye

i think 16:10 is WQXGA

so 2560x1600

[split steppe]

ye

actually not 30" i think 24"?

or 26"

something like that

i forgot the size that made sense

[olive sable]

iirc they come in 30 and 27

[olive sable]

3D printers started using 16K actually

[split steppe]

3d printers make sense, you want as much res as possible

looks like the 30" is the right one, can't find any smaller wqxga, and it seems to be the one dell i looked at

sadly they don't manufacture that size anymore

[proven merlin]

And i am once again stuck

I cant get opnsense or whatever to give my debian vm a ip

Dnsmasq dns and dhcp are on

Theoretically set up aswell

Unless i missed something

But the debian vm just wont grab a dhcp ip.. i can set one manually then talk to the opnsense gui

But internet is also dead

[split steppe]

possibly the vps provider enforces something from outside

[proven merlin]

They shouldn't be able to?

I mean im just running a vmbr internally

[split steppe]

well your vps is just someone's vm, which they integrate in the rest of their net infra

[proven merlin]

The public ip to opnsense works, but the opnsense to vm net stuff ain't working

[proven merlin]

Yes, but they cant look into whats in my machine.. to them it looks like one machine sends quite a few Ethernet requests..

[split steppe]

hmm perhaps

[proven merlin]

Its struggling internally

[split steppe]

i suggest you draw a diagram on paper

[proven merlin]

The opnsense can access the net aswell

[proven merlin]

Probably should, or probably should also go to sleep lol

[split steppe]

yuh

[proven merlin]

The fun part is ONLY opnsense can access the net..

My proxmox and my debian (which sit on the vmbr behind opnsense) cant

[split steppe]

something you should also check is SELinux settings, it's prevented my network config from working sometimes

then firewall and then ipconfig

try with fixed ips as well

[proven merlin]

To be fair, my proxmox uses a very uh.. interresting mac.. 02:00:00:00:00:01
But i cant see why that would cause issues

[split steppe]

doubt mac matters

it's just device identification

[proven merlin]

Proxmox Firewall is disabled on all interfaces.. even enabled promiscuous mode in opnsense

[split steppe]

proxmox might be controlling your whole net config also

since it needs to bridge the vms and stuff

[olive sable]

i might switch from vllm to llama.cpp.
mainly due to llama having the ability to pause the llm processes and to temporarily save the llm model in ram while the gpu gets used by something else.
idea now is to move tailscale into a docker container, have nginx as proxy to acces localhost services, and to have slurm to schedule the GPU

i think this works

might need to make a ramdisk to save the llm into while its paused

[split steppe]

with llama.cpp you can configure specific offloading configs between disk, ram, cpu, gpu, you shouldn't need a ramdisk

[olive sable]

its not about offloading

[split steppe]

mmm

[proven merlin]

My interfaces config is fairly simple.. but i do have an ip + gateway setup on the vmbr1.. which should be fine.

[olive sable]

i want the llm to be online 24/7 if the gpu is free, but if im training osmething on the server then the llm must go

but just killing the processes is annoying and it will take 10 min to restart

[split steppe]

hmm

[olive sable]

so idea is to move model weights from VRAM into RAM until the gpu is free again

[nocturne olive]

[olive sable]

and then back into VRAM

[nocturne olive]

Surely that's gotta be possible somehow

[split steppe]

would ramdisk even do anything for that though

you'd still need to rebuild the entire program state

[olive sable]

hmmm

reloading from ssd everytime would nuke the ssd tho after some time

[split steppe]

much of the time spent is packing weights into the gpu/ram in a specific way and initializing the program

[split steppe]

reading doesn't really decay the ssd

[nocturne olive]

I mean Linux does copy the whole model from disk to RAM first before actually loading it in llama.cpp at least
So if it's in a RAMdisk you'd get the benefit of near-instant copy so reloading the model would be way faster if it was in a prepared RAMdisk

[olive sable]

thats what im thinking ye

[split steppe]

it doesn't even, by default it mmaps

[nocturne olive]

[split steppe]

so if you use a MOE you don't even load 90% of it

[nocturne olive]

In my testing of big files, when you load a big file EXT4 first loads it into RAM cache, then lets the app load it from there

The app only gets notified of the transfer once it's completed loading to RAM cache

This does however lead to some stuttering sometimes while games load

[olive sable]

im pretty sure its loaded tho. vram is full and system ram is only at 8gb

[split steppe]

that's implementation dependent, there's a huge diff between a full load and mmap

depending on what type of model you have

i run the default mmap mode and it's a difference between 30s startup and 3+ minutes

and massively different ram use

no execution speed diff

[olive sable]

i don't care about ram usage at all, i have 256gb so enough to spare. but fully starting vllm takes 10 min from when i enter the command until the llm starts workign in the webui

[split steppe]

what size model you have? it could be doing a full load instead of mmap

not sure what vllm defaults are

also is it dense or moe

[olive sable]

Should be 16GB with vllm just hogging the rest of the vram cuz its greedy

gemma-4-26B-A4B-it-heretic-AWQ-4bit

[split steppe]

yea that should load in like <1m

[olive sable]

[split steppe]

my qwen3.6 35B-A3B does

[olive sable]

It takes a while for me

[split steppe]

granted i'm running on cpu, but i don't think gpu should be that diff

[opaque sigil]

they might be JIT compiling a bunch of things during startup + capturing cuda graphs alongside warmup and whatever else they do can take a while i guess

copying 16gb takes a second maybe

[split steppe]

~40s to boot, with repack and warmup

22G model

maybe some cuda shit is slow

[olive sable]

Iirc it does some testing or whatever and something about cuda graphs

[opaque sigil]

try with --enforce-eager sam, wonder how much that helps with startup

[olive sable]

Hmm

Sure

[opaque sigil]

granted serving will probably be slower

this basically just disables cuda graphs it seems

i don't have too much experience with cuda graphs since most things i deal with don't work with them but iirc massive graphs can take a while to compile

[olive sable]

hmm

[opaque sigil]

surely there's a way to have vllm log everything it's doing during initialisation and for how long

[olive sable]

i have this:

ok maybe not 10 min, but it takes longer than id like

roughly 5 minutes it seems

[opaque sigil]

almost a minute for torch.compile

[mighty thorn]

vllm

[olive sable]

another case of x group hating users of y software for no reason

[mighty thorn]

Either use Ollama or use that brain of yours to make ur own inference api

vllm is in the middle of those and is the worst of them both

[olive sable]

based on what?

[mighty thorn]

All of the bloat of Ollama with all of the technical difficulties of diy, talk about a lose lose

[olive sable]

i dont have either tbh

[frozen igloo]

So close yet so far

[frozen igloo]

https://youtu.be/nLu4n7yNGdk?si=qxd0ASEYprXpx3Yq
cool thing

I would love to see a much more thorough version of that with muscles, neurons, and soft body physics

[stark needle]

I keep getting this shit recommended as well

https://blog.google/products-and-platforms/platforms/android/gemini-intelligence/

Vibe coded slop design

[young plover]

I only used google assistant to set laundry timers and I stopped doing that even

[stark needle]

Oh my fucking god

They really made

Google ai slop copilot laptop

https://blog.google/products-and-platforms/platforms/android/meet-googlebook/

Chromebook with Gemini (cloud Version)

Surely we didn't have this before

[frozen igloo]

The next step is a real-time ai generated operating system. If your computer can’t forget that you had a window open then are you really computing?

[leaden crest]

I have a bit of money riding on at least one tech company anthropomorphizing alexa/siri/google assistant within the next half decade

[worthy niche]

does project ava count?

[leaden crest]

I saw that actually and im not sure honestly

wait the design similarities to nere are actually uncanny

now that I look closer

[worthy niche]

im sorry what

i need a side by side

[leaden crest]

mmm maybe im blind idk

yeah im blind

[mighty thorn]

Ai Minecraft for windows

[leaden crest]

nere if someone actually tried on the model

instead of going for simplicity

[worthy niche]

i was gonna say ai windows but they beat me to it

windows is already ai

why am i productive wrong
i keep getting distracted programming when i should be getting my grades up

[marsh coral]

all these big tech corpos shove AI into random things

also gemini is mid, closed source, and has google tracking, soo

[marsh coral]

theres already an OS with AI in its system level

i think its

Deepin

they have this thing called UOS AI

deepin is a chinese linux distro

UOS AI is basically copilot but its embedded into ur OS

instead of being scatterred across apps

and u can use a local model or bring ur own API key for UOS

[frozen igloo]

Nothing could possibly go wrong

Is it able to delete files?

[marsh coral]

i think it depends

its defo doesnt have root/admin access

[frozen igloo]

Oh good lol

[marsh coral]

its mostly a chatbot/assistant

so u can summon it to write docs or emails or a summary

and help schedule stuff

it has "claw mode" which is basically the mode that makes it an agent

so it executes commands

in the terminal

[chilly drift]

Hello

[frozen igloo]

Hello

[olive sable]

hi

[chilly drift]

i don't know what to talk about I just wanted to see late night programmer talk

[marsh coral]

its like 11 AM for me tho

[chilly drift]

Oh wow

Other side of da planet

[marsh coral]

im in indonesia

[chilly drift]

Ooo
Cool
I'm in the USA so

[marsh coral]

lemme guess

west coast?

california

[chilly drift]

Hahahahah No

Very far off actually

[marsh coral]

uhhh

east coast?

or texas

or somewhere in the rust belt

[chilly drift]

Well, the opposite of West is east so

[marsh coral]

i meann idrk abt american timezones

[chilly drift]

I'm from the deep south
But not so deep it's Texas

[marsh coral]

mhmm

okk

[chilly drift]

I like this guessing game

Glad my accent doesn't come through text Haha

Some people were talking about ProxMox servers earlier. I was reading up

Does anyone have advice for using ProxMox?

[marsh coral]

whats that?

[chilly drift]

Proxmox is an open source server management OS with a web based interface

[marsh coral]

hmm ive never rly heard of it

[chilly drift]

It's mostly used to manage VMs, LXC containers, networking
Stuff like that

[marsh coral]

ohhh

ok2

[chilly drift]

Lots of people here were talking about it
You can search it up if you're really interested.
It's becoming more and more popular since it's completely open source

It's fun to play with especially since it's a quick and easy way to make isolated VMs

Meaning you can mess around with whatever and not really worry about breaking literally everything in your network

[marsh coral]

truee

[glad path]

undertale lol

it makes sense?????

that genuinely makes sense though

wym why i it default

[frozen igloo]

How on earth would you go about loading an entire game onto ram? Would you need a mod? I remember in the assassins creed stream, vedal had enough ram for the whole game and i think someone suggested that

[glad path]

you could load undertale into ram

[frozen igloo]

That would be a funny performance mode for people with an absurd amount of ram but no ssd for some reason lol

[frozen igloo]

Running the game takes roughly 10x more ram than the game itself
According to the requirements

[patent shard]

that's not too surprising

[glad path]

ok

the game is like

200mb

[frozen igloo]

That’s what I was referring to. I just thought it was funny

[tight tinsel]

all this when i just use glue and a dovetail 💔

man i stink at 3d design

[rigid snow]

??

it looks great

no taste

[true hemlock]

shir

[stark needle]

✅

[amber fractal]

If you alone somehow managed to gain control of mana, the fuel of all magic in this current world. Assuming that you can regenerate mana over time, what would be your goals to do with it?

[stark needle]

[true hemlock]

make gold

[amber fractal]

I feel like making gold is a safe bet, who knows what magic system will be used. But most of them at least have that as a possibility

[olive sable]

My 3D teacher is subbed to acerola

[stark needle]

Who is acerola

[olive sable]

Youtuber

[proven merlin]

Acerola mentioned

Peak

"But Acerola"

[proven merlin]

Ram disk, aka Turning ram into a disk drive and then copying the game over. Either via TMPFS or ZRAM on linux ig. Windows has some too

That would be the easiest solution

Would also use about 2x the ram of just the game off of disk

[trim valve]

thanks amd driver update

very reassuring

[hard delta]

inb4 it's an update from windows update 🤭

[trim valve]

...no?

but whatever makes you feel happy dude

[jagged turtle]

did you even read what was being said in the text accompanying the image?

[hard delta]

whql updates pushed by microsoft are the ones that usually do weird crap

yes, when microsoft pushes an amd driver update it's as much an "amd driver update" as when you fetch it from the site

i can read

[jagged turtle]

I'm still very confused

[trim valve]

ah

that

would be why my firefox is complaining about disk usage

[jagged turtle]

slight chance (how does that even happen)

[trim valve]

no idea

[hard delta]

maybe that one was me too

[trim valve]

fixed after restarting firefox

[fickle rain]

Can you use paged pool for ramdisk? 🤔
IG if you wanted ultra optimized you’d wanna avoid page faults

[silent cloak]

With how little disk space i have i should do this

I have more RAM than storage currently

[flint dome]

the error is because the file is saying amd ow

the amd hurt

[upbeat cradle]

https://cdn.discordapp.com/attachments/512346452148748289/1503516847406845953/live_capture-ezgif.com-optimize.gif Gave Max real eyes finally. Now he is slightly more creepy.

[silent cloak]

https://tenor.com/view/wawa-emdeb-fail-failure-cat-gif-16703817715131368982

[stark needle]

Catman

[stark needle]

Egg cat

🥚

[glass jetty]

https://tenor.com/view/neurosama-gif-15292687851529300845

[stark needle]

🦊

[stark needle]

https://tenor.com/view/my-beloved-gif-16745057618688160042

[glass jetty]

https://media.discordapp.net/attachments/870411388475031572/1497906625036550275/caption.png?ex=6a03a901&is=6a025781&hm=31d97eb1849ae9aae013ddd087b17839aefd6089a73e8c4a502519fde8b747ca&format=webp&quality=lossless

[stark needle]

🦀

Crab

Openclaw

[stark needle]

[stark needle]

[glass jetty]

btw

i not even remember what i post

it just random letters

oomg 6 7 ok

i wonder what else i have

[glass flower]

please someone tell me the original song for the 6 7 animation meme... i can't find it

[glass jetty]

omg ok
build finished with error
databasecontroller.cpp:184:20: error: use of undeclared identifier 'testingMaterialChange'
testing_call(testingMaterialChange(materials));

right

[glass flower]

[stark needle]

:nrooroCarUbhh:

[jagged turtle]

what was just deleted

[glass flower]

i'm still looking for the 6 7 song.. i got a earworm and i need to listen to it on loop

[silent cloak]

https://tenor.com/view/no-nope-no-way-noo-absolutely-not-gif-20244171

[jagged turtle]

wtf

[glass jetty]

i really have just random numbers and nothing else

[glass flower]

i found it...

https://www.youtube.com/watch?v=55NduFO4_vM 6 7

[flint dome]

the_sun_is_racist.mp4

[silent cloak]

L organization

Sort by categories

[glass flower]

didn't know this channel got renamed to #1194088101627297924

[stark needle]

Is this #1194088101627297924 now

[glass jetty]

yes

active

unless some one start new ai topic

[flint dome]

"this goes into brainrot"
"this goes into mlg edits"
"okay what the fuck even is this"
"this one goes into dark humor"

[stark needle]

https://tenor.com/view/rule-rule1-brick-wall-talking-to-a-wall-argument-gif-22291988

[silent cloak]

Literally what ive done for decades

Then i got a reaction image folder

[glass jetty]

may leak something at this rate

[silent cloak]

And in there i got a neurosama folder

[glass jetty]

its all just downloads folder

[stark needle]

I remember the spanish mlg videos I'd use to watch

[glass jetty]

i dont remember what i already posted and what not btw

[stark needle]

Today in 2026 that creator would get cancelled 1000%

[glass jetty]

stuff i downloaded from discord to upload in tg

[flint dome]

my meme folder is too scattered on twenty hard drives to bother sorting so its all sorted by time period automatically

[silent cloak]

Someone needs to post new slop models like gemma or qwen

[silent cloak]

Couldnt ever do that because of naming when i needed something specific

[stark needle]

https://huggingface.co/closedagi/gpt5-r3-claude4-magnum-dong-slerp-raceplay-dpo-4.5bpw-1.1T-exl2-rpcal

✅

[silent cloak]

Lmao

Literally every ai model post in here

[flint dome]

magnum dong

[jagged turtle]

I tabbed out for 5 minutes why am I in #1194088101627297924

[flint dome]

approved

[jagged turtle]

what in the fuck is this model name

[flint dome]

I personally call it the long kong

[silent cloak]

Literally beats every qwen and gemma model

[glass jetty]

there another portion

since chat died again

[silent cloak]

I cant send 99% of my memes

Especially on this account

[glass jetty]

sadge

its kinda isnt even my memes

it just stuff i downloaded to share

[silent cloak]

Good memes are a rare commodity these days

If its from reddit i shred it

[silent cloak]

2016-2018 by the looks of it

[glass jetty]

oldge

there no even vibe coders there

[silent cloak]

They arent real parts of IT

Its more of a cosplay

[glass jetty]

nicest thing to see is that several (3) your abstractions each tested separately and designed for different task
fuses without ANY errors for the first time (after like 3 hours of compilation errors but anyway)

[silent cloak]

Lmao

[true hemlock]

ddr6
https://www.geeky-gadgets.com/steam-machine-delayed-ram-crisis/

[stark needle]

Is that shit coming out as well now

Bruh i havent upgraded to ddr5 even

[glass jetty]

me too

[true hemlock]

this feels more like ai slop article

the reason i said ddr6

is because they mentioned ddr6 everywhere

[glass jetty]

lets see if it is slop

[true hemlock]

"valve delayed steam machine because ddr6 shortage"

[stark needle]

Supposedly ddr6 is 2027

[true hemlock]

[stark needle]

Desktop 2028-2029

[stark needle]

Maybe they meant gddr6 idk

[true hemlock]

gddr6 isn't in shortage

[glass jetty]

100% everywhere

[stark needle]

Why dont they just make fucking GPUs with that then

[true hemlock]

we already do

amd gpus

and nvidia going back on 3060 prod

[stark needle]

Bruh is that why so much cheaper

[glass jetty]

isnt there ddr7 on gpu

[stark needle]

Gddr7 is nvidia slop

[glass jetty]

ok

[rigid snow]

aa

[silent cloak]

How we gonna get ddr6 with the current bs

[nocturne olive]

Me last night

[true hemlock]

imaginary

[silent cloak]

Placebo ram

[mighty thorn]

“Technology after 2020”
“Is this… slop?”

[split steppe]

it's qualityslop, you just like it because it's good

[mighty thorn]

So true

[kind nimbus]

TIL this is a real book

[opaque sigil]

bzzzzzzzzzzzz

[kind fable]

wait that's Jython ?
https://fxtwitter.com/akramcodez/status/2054099010571645430

Sk Akram (@akramcodez)

Use whatever brace style you prefer.
︀︀
︀︀But not this.
︀︀
︀︀Don't do this.
︀︀
︀︀Seek help instead of this.

**💬 163 🔁 307 ❤️ 3.8K 👁️ 133.4K **

FxTwitter · May 12, 2026, 7:19 AM

[kind nimbus]

https://cdn.discordapp.com/attachments/1352269813606649926/1352269813900378122/Neuro_Sunflower.gif

[stark needle]

✅

[rigid snow]

that's pyava

[nocturne olive]

What the heck

[kind fable]

I hope they align the trailer with tabs

[stark needle]

🪹

[kind fable]

Bruh wtf is that Jetbrains

[kind nimbus]

https://cdn.discordapp.com/attachments/1072697081443131476/1366110037193392259/zoomieSmol.gif

JetBrains redeemed frog hat

Now with extra less brains

[rigid snow]

extra less

[stark needle]

🪹

[rigid snow]

empty_nest

Click to learn more

[stark needle]

@gokr

is this true

[rough bloom]

@silent anvil

[kind fable]

guys new Linux LPE just dropped
https://github.com/v12-security/pocs/tree/main/fragnesia
https://github.com/user-attachments/assets/d8cdf3ad-2874-4a92-9a2e-46ae6e9a6761

[glass jetty]

https://github.com/robertdfrench/ifuncd-up

free $500

[floral hawk]

I'm buying a second GPU, with it I'm gonna be able to diversify. Zero will get a much more powerful model AND she'll get a seperate model entirely for gameplay + dedicated vision running on VRAM and not RAM.

[nocturne olive]

What GPU?

[floral hawk]

Its a 1080 a buddy of mine has

All it has to handle is tool call and vision, should be fine

I want to also possibly run OBS off of it, maybe.

[nocturne olive]

1080 isn't supported by latest Nvidia driver
What GPU do you have right now?

[floral hawk]

But I'm not sure how that would play into my display setup

4070ti

It's really just for rendering. . . man its out of support? Yikes

[nocturne olive]

Yeah na your GPU driver won't support the 1080 probably, you'll need to downgrade

Max driver version is driver 580

[floral hawk]

Dang, I can find another one

[nocturne olive]

Anything 2000 series and up is still supported by the latest driver

[floral hawk]

What about the 3050?

It's 6gb and will serve the purpose I need it for fine

[nocturne olive]

The 3050 is supported by the driver
Also the 3050 should have 8GB, the 6GB one is a scam and you should not buy it

[floral hawk]

Only reason is someone is selling one on FB marketplace for like 50$

Im already aware of the 3050 bs

[glass jetty]

ppl still use 700 and they just no longer provide drivers...

[nocturne olive]

[glass jetty]

[kind fable]

I wonder

[glass jetty]

im not

[opaque wharf]

750Ti my beloved

[glass jetty]

true

[kind fable]

Could we train neuro to sing french ?

[glass jetty]

no

no

no

[kind fable]

But

[glass jetty]

NO

[kind fable]

I know she can talk french i remember one collab where she did

So surely we can make her sing

I swear there are beautiful french songs

And actually it goes for all languages

[opaque wharf]

La Mersailles - Neuro Sama ed

[kind fable]

I wish neuro could cover nine nine luftballons

[glass jetty]

no french

[nocturne olive]

Do you have labaled French singing?

[kind fable]

I could

[opaque wharf]

Neuro singing god save the king

[nocturne olive]

If we have data, we can add French to NeuroSynth's capabilities

Cross-training

[kind fable]

Alright how do i need to format the data exactly

[nocturne olive]

You need to record a bunch of French singing, then use the French phoneme set to label it using a tool

[glass jetty]

i shouldve keep you muted

[nocturne olive]

@high brook Go instruct them how to data in DMs or something
Unless you want me to

[kind fable]

I want Neuro to do songs in more languages cause it'd be cool

She already does swedish

We can do more

[nocturne olive]

Neuro's Swedish isn't actual Swedish proper though

[opaque wharf]

Carameldansen

[nocturne olive]

It's using phoneme trickery and kinda half-sucks

[kind fable]

Sadly

[kind fable]

Sadge

Dota is fire

[nocturne olive]

I'm planning to add native Finnish to NeuroSynth 2 tho

[kind fable]

If even with a trickery which half sucks we can get cool stuff then with proper stuff

[nocturne olive]

I know of exactly one UTAU that sings actual Swedish (it's peak)
The same UTAU also sings Finnish

[kind fable]

Oh wait

Making the training data is basically like making an UTAU set ?

[nocturne olive]

https://www.youtube.com/watch?v=besGCxz-WNE
Did I do that right

[high brook]

I'm alreadying dying trying to get german working

[nocturne olive]

Well here we have someone we can borrow French from purpose-built for our needs

[kind fable]

Wait we could get nine nine luftballons

Yes yes yes

[high brook]

THATS THE ENTIRE POINT

[nocturne olive]

[high brook]

THIS MAN GETS IT

[kind fable]

YES

[high brook]

that and ode to joy

[nocturne olive]

[high brook]

so many good german songs

[nocturne olive]

Well if you want to get us data then go ahead and ask Wispers for the data creation instructions

Wispers is our data specialist

[high brook]

somehow

[kind fable]

I go ahead

[high brook]

It's mostly me being stubborn

[nocturne olive]

I just put it all together and sometimes do some labeling

And kinda lead the project I guess

[kind fable]

Good good i'm that too

[nocturne olive]

And then do some fun stuff

[kind fable]

Once i enter hyperfocus mode i'm. On a task for a while

It's called the ADHDtism power

[high brook]

I'm also the guy that does tech prototyping

[nocturne olive]

And I do tech proper

[kind fable]

I should not fall into rabbit holes
But i still do and it's fine™️

Oh ping

[high brook]

RN I'm trying to test a new backbone, but got stalled by a bug the dev ddint even know existed till just 15 minutes ago

(I reported it)

[kind fable]

Don't we love finding bugs

[high brook]

I've reported so many bugs to this person.

[kind fable]

Find a bug : I'll report it later
3 years later : i hope someone fixed it
1 year later : i should maybe send my personal patch

[nocturne olive]

Meanwhile me finding and fixing a memory leak in the tooling

[kind fable]

Solve it by RiR /j

[nocturne olive]

And implementing CUDA into a different tooling

[high brook]

in this case someone had changed a variable name in the backend tech stack and nobody noticed

I came up with this funny sentence though

[kind fable]

But yeah send me instructions for how i can do voice in fr*nch

[nocturne olive]

New slave labeler

[high brook]

probably someone shatardizing variable names

[kind fable]

Neuro is my Queen, i pledge allegiance

[nocturne olive]

What kind of voice type may you possess?

[kind fable]

I'm not mute

[nocturne olive]

I was thinking like gender and accent type

[kind fable]

And i'm also not deaf

[nocturne olive]

Different voices have different suitabilities, even if all can basically be used

[kind fable]

I'm less than optimal

My voice is quite far from what one would want from Neuro or Evil
I'd be closer to Vedul

[kind nimbus]

Otherwise you couldn't hear evil's karaoke now

[nocturne olive]

As long as it can contain all the required phonetic information cross-training should sort it out

[stark needle]

Is there no phonemizer u can use

Surely must exist

[kind fable]

Technically i was the best singer back when we had a singing class
I also didn't have my adult voice yet

[nocturne olive]

Training is completely disconnected from phonemization

Well outside of training using phonemes as labels

[kind fable]

Yay 42

[proven merlin]

Just be aware to not turn off your pc or store important data and hope to have ecc

[kind nimbus]

What happened there on twitch?

Html code leaked? Lmao

<fim_suffix/>

[proven merlin]

Not entirely sure what paged pool is.. i know you can make a vram disk. For whatever thats worth

[kind fable]

Don't worry
It's called a bad bytecode patch

[opaque wharf]

VRAM disk...

[urban lance]

what project do someone recommend for me to learn zig

[proven merlin]

Because im me, ill say ansi c compiler

[maiden geyser]

hello #baking

[amber fractal]

[mighty thorn]

first actual ai solve

[stark needle]

Write rust compiler in zig

✅

Then move to rust

✅

[amber fractal]

[stark needle]

✅

[urban lance]

thank you but i want to live

not to die from depression

[fiery canopy]

we need

smthin

a taste of drama

we need this channel poppin agian

or sm shit

sm1 say smthing

[mighty thorn]

I can do that

Easy

People’s reactions to the Meta Raybans smart glasses with the cameras are fucking stupid

The glasses have a recording light, phones don’t

It’s easier to sneakily record someone with a phone than those

[fiery canopy]

I think the main issue people have is removing the light

but

still stoopid

almost defenitely can record with many other means

the glasses are just easy

[mighty thorn]

If your so scared about the chance of being recorded without your knowledge, bad news about the last 15-20 years of technology

[fiery canopy]

this is the same reaction

They have to anything tho

[amber fractal]

[fiery canopy]

is the best way this channel stays active

lmfao

[amber fractal]

I want to build a AI image generator

[fiery canopy]

atomic strike incoming

[amber fractal]

[mighty thorn]

isnt that already what ur pfp is

[amber fractal]

That's yours

[mighty thorn]

what

[kind fable]

Don't worry it's quite simple
for x, y
Random.integer

[hard delta]

imagine a world where we can get infinite entertainment generated... nobody has to pay for music or tv anymore and its always a unique, different render!

[fiery canopy]

I think I succeseeded in reviving chat

[amber fractal]

My PFP is not AI generated

[fiery canopy]

time to mysteriously vanish back into the shadows

[kind fable]

Was it ever dead ?

[amber fractal]

No

[fiery canopy]

for abt 30 mins

:0

too long for me

[kind fable]

No it was just in hiatus

[amber fractal]

Go to NN then

[fiery canopy]

ehh

[kind fable]

While i'm learning how i'm gonna make neuro sama sing in french

[fiery canopy]

NN too much brain damage

[kind fable]

It's both more complex and simpler than i imagined

[fiery canopy]

sounds abt right

[kind fable]

We are not a cult right ?

[amber fractal]

Not yet

[warped narwhal]

Not with that attitude

[mighty thorn]

Can’t tell if your being or just baiting into further debate about ai image generators

[amber fractal]

I think AI image generators are pretty cool

[hard delta]

this was a starry eyed, bright future agenda ten years ago

[mighty thorn]

Pretty funny that people wanted ai until they realized that it needs stuff to learn from and hardware to do learning on

“I want this but it’s not allowed to have any consequences or aspects I dislike or else you are worse than Lucifer himself”

Literally “make infinite entertainment, make no mistakes” prompting speak

Dreamers were the original vibecoders

[hard delta]

yeah people expected it to be a benevolent 200iq mother running on a single potato

plus even today their main issues are corporate, not technological

[mighty thorn]

“Make an intelligent machine that knows everything and can draw but also don’t give it any information or a pencil”

[hard delta]

well duh if it's so intelligent it should divine all the data from its big intelligence! isn't that what geniuses do?!

[amber fractal]

What is human processing

[rigid snow]

genai in art is acceptable

but only if i like it

this make sense?

whatever i retreat i am arguing with myself now

disregard

[kind nimbus]

https://cdn.discordapp.com/attachments/1067638175478071307/1486469601629044957/dsdsqfsdqfsqdfqdfqdfqfd.gif

[rigid snow]

ok i can't leave this now. what i was thinking is basically it's fine if it's to a reasonable degree. if a sample a song is built on is ai generated does it matter? it still made the artist feel something. if img2img was used in making digital art does it matter? if the artist felt something and decided they intend to put that in because it resonates with them

my point being it doesn't make it not art magically

though where i'd draw the line is i'd say crafting a prompt isn't a creative process in itself

more accurately it is, but the result isn't the product of it

[hard delta]

honestly this thing happened with digital art and Ctrl+Z too

[worthy niche]

yeah, if you have an ai generate the entirety of your art, that isn't your art. You didn't make it. Your thought process and soul were never a part of it

[rigid snow]

bad analogy i think, i wouldn't draw parallels with digital art as a whole

[worthy niche]

anyway back to doomscrolling

[frozen igloo]

https://tenor.com/view/eu-gif-18523165

Ctrl c, ctrl v

[rigid snow]

basically yeah to me it's not about the entirety or a threshold it's more about how involved you were and how much you've put yourself into it

i feel like the common opinion rn is if i used ai generated salt on my dish it's suddenly slop and i'm not a cook

[hard delta]

free lunch slop

[rigid snow]

and is it a bad thing i contradict myself? i can't/feel horrible using ai in crafts i'm good at, it almost hurts my ego if that makes sense

[mighty thorn]

Being of slop, have some opinion slop!
These are on a video covering a research paper using ai to map global farmland.
The anti-ai parroting has penetrated so deep into their minds that they are rationalizing this as a bad thing

[rigid snow]

wait, is it an ego thing? did i just crack it

[frozen igloo]

Everything is “dystopian” now

Every cool technology thing is actually terrible because “the military is gonna use it”

[rigid snow]

the only ai comment i see is "haha ai water" and even then it's just because ai was mentioned

[mighty thorn]

Dude my history teacher just docked points off my exam because I made a spelling mistake.
Think about how dystopian that is dude. The history teacher thinks he has a right to dictate how I spell words

Dude

[rigid snow]

also wysi

[mighty thorn]

You are aware people can talk about things without explicitly mentioning the thing in every sentence?

[rigid snow]

it just as well could have been comments about the exact same thing but done manually

the common topic isn't dogpiling ai, it's surveillance

[mighty thorn]

We both know they wouldn’t have a problem with it if it was framed as the end of a 15 year citizen science survey based mapping experiment

The issue is clearly those two scary letters that everyone needs to hate

[rigid snow]

well if it was framed as something helpful yeah you'd get less of that

trust me i don't like the anti everything ai crowd too but here it's just not related

[mighty thorn]

Most people only hate ai now that it’s good. Making me think that, perhaps only subconsciously, it’s out of fear

No one really cared about original ChatGPT 3.5, people got pretty mad over stable diffusion, video generators were ignored until Sora 2, etc

[rigid snow]

hating ai is valid for a handful of reasons, but people lack nuance nowadays

sd 1.5 based checkpoints are still really good even today

[mighty thorn]

So true

[hard delta]

i'd say in some cases it's job security, and in all remaining cases it's confounding the tool with who uses it and for what

[mighty thorn]

Anti ai people are actually even better than ai companies at picking the worst possible option when presented with multiple

1. “Ai steals from artists”
2. “Muh water”
   option 2 selected

My face when heat is used to evaporate water which is then released as steam (I am fucking horrified by the water cycle)

[rigid snow]

to be fair the water thing isn't invalid, it's an actual issue. the problem isn't that the water disappears because it doesn't, it's because it's fresh water that's used to cool shit

and when it eventually falls down as rain it's not a guarantee it ends back up in a fresh body of water

relevant mostly to dry regions

[mighty thorn]

I think the actual bad actors here

Aren’t the corporations (directly)

Or the anti ai activists

Or the country politicians

It’s the county mayors just approving stupid datacenter projects for money

[rigid snow]

the actual problem is that we lack bigger problems

hantavirus boutta fix that one 🔥

[mighty thorn]

“Hey sure you can build a datacenter on that orphanage which uses 11x the power output of our entire grid, just gimme $5 first”

[woven temple]

I think AI is neat tech and yet I think it will push humanity as a whole further into a dystopia.
Billionaires getting richer while everyone else looses jobs.
In an ideal world people should be able to create art and culture because AI does the hard and tedious work. AI creating art is not empowering anyone, it is feeding lazyness and devalues emotion.

[high brook]

Be glad is an Argentina strain and not a Korean one. The Korean strains are terrifying

[mighty thorn]

Ehhh

There are legitimate uses for ai art

Like placeholder assets

References for making real art

Etc

Also some editing tasks like greenscreenless background removal

[tiny edge]

nerds of this chat, what do yall think of medal(the software)?

[mighty thorn]

Explain in exactly 8 words so I don’t have to google it

[tiny edge]

clipping software with extreme poor performance some use

[mighty thorn]

I think you should pick one with extreme high performance

[tiny edge]

OBS

[ivory tiger]

it's like having a console for recoding gameplay, nun serious but perhaps gets the job done for casual purposes

[sick owl]

For those of you guys who were looking for a better alternative to lmstudio

https://github.com/oobabooga/textgen

textgen webui is now a self contained desktop app shipping with ik_llama.cpp builds

[leaden crest]

https://thinkingmachines.ai/blog/interaction-models/
have we talked about this in here yet

seems pretty cool

not to mention neuro related

[mighty thorn]

Aren’t those the guys who beat ChatGPT at ARC AGI with 27m parameters

[leaden crest]

think so but dont quote me on that

there’s no public data for this available yet but it seems legit so far

[blissful geyser]

i also have a concept idea

you know how there's open source models of voice generations that aren't limited by tts pronounciations

wait that's just evil neuro nevermind

sorry brainfog

[kind nimbus]

[mighty thorn]

What

[blissful geyser]

nothing just coping about the v3 voice

[amber fractal]

Tiling images for sampling data

I'm waiting for people to realize this is not a meme and scream from the hills to get me banned for AI art promotion

[opaque sigil]

i don't really care about AI art personally as long as i'm not the one who has to implement it

i want nothing to do with the math behind diffusion and all the other generative models

[amber fractal]

I'm just slopping the model and surely it works

Single CPU core you can crop the 10 images 40k times

[sick owl]

A guy found an innovative new approach to cooling his DGX

[sick owl]

Oh that's extremely cool actually

Didn't know thinking machines were finally making moves, thanks for the heads up

[kind nimbus]

Just strap some big ass fans to it

[sick owl]

This logo offends my eyes

[kind nimbus]

They have some nice fans

[patent shard]

they make lights, too

[sick owl]

Bruh they removed the 80% charging limit toggle in a GNOME update and now my laptop is stuck on it if I don't want to go digging for a fix

Not that I planned on disabling it but still

[sick owl]

Very cool if it holds up to their demo

https://www.youtube.com/watch?v=A12AVongNN4

This is the big one

[opaque wharf]

Lmao, everyone but OpenAI are innovating

[kind nimbus]

Lmao

[opaque wharf]

It should be somewhere in /sys/batt/batt0/charge_end_control_threshold

[sick owl]

I downloaded the wrong installer and couldn't be arsed to rebase

It is what it is

[sick owl]

Assuming they're not faking it with a tool call on the background model

I'll give them the benefit of the doubt until its out to test

[kind nimbus]

Then you got GNOMEd

[sick owl]

https://giphy.com/gifs/Enr4bc3JeOOzHLYY8R

[hallow dagger]

https://tenor.com/view/coding-what-does-boolean-mean-gif-11550719031942041980

[frozen igloo]

Now I want a roomba that says friend every time it sees someone new in the room it’s in

[sick owl]

I can only imagine what a nightmare this was to engineer

This is impressive if it holds up to scrutiny outside of benchmarking

[opaque wharf]

Maybe neuro could be improved further

[sick owl]

https://vxtwitter.com/thinkymachines/status/2053938897265344737

Thinking Machines (@thinkymachines)

↪️ Replying to @thinkymachines

Our model interjects to save Alex's parents from his eccentric ideas.

▶ Play video

vxTwitter / fixvx · May 11, 2026, 8:42 PM

[sick owl]

My hope is given how open this research is we start to see open source replications crop up

[opaque wharf]

True

Because I think this is currently quite the innovation in LLM space

[sick owl]

https://vxtwitter.com/thinkymachines/status/2053938899735724254

Thinking Machines (@thinkymachines)

↪️ Replying to @thinkymachines

With the model's simultaneous speech capability, Horace has gotten a lot easier to work with recently.

▶ Play video

vxTwitter / fixvx · May 11, 2026, 8:42 PM

Some more demos

Here's the time awareness video

https://vxtwitter.com/thinkymachines/status/2053938901438681184

Thinking Machines (@thinkymachines)

↪️ Replying to @thinkymachines

The team has been sweeping at local trivia night thanks to a model that's aware of continuous time.

▶ Play video

vxTwitter / fixvx · May 11, 2026, 8:42 PM

And the last one

https://vxtwitter.com/thinkymachines/status/2053938908296315060

Thinking Machines (@thinkymachines)

↪️ Replying to @thinkymachines

Lili and Martin get some help controlling themselves.

▶ Play video

vxTwitter / fixvx · May 11, 2026, 8:42 PM

I think the thing that really impresses me is its apparent proactivity

I haven't seen a system behave like that before without serious prodding to do so

[sick owl]

Stretching the definition of "instantly reacts" though

That said even with those flaws, its still a very promising proof of concept I'd say

[opaque wharf]

The famous Neuro negative latency

[leaden crest]

true!

[leaden crest]

my first thought too

[nocturne olive]

I have an almost perfect new remote desktop setup using Sunshine and Moonlight-qt but for some stupid reason the mouse area offsets are wrong so the mouse ends up being placed at the absolute coordinates relative to the X11 display root rather than the display's coordinate space meaning on my left low down 1080p monitor I can't reach the bottom and on my 1440p monitor on the right I can only reach a bit of that monitor and most of the area maps to the 1080p screen's area

[leaden crest]

vedal’s been against a total neuro rewrite (codebase doesnt count it just migrated languages) but maybe this will inspire him to make a new architecture or smth

[leaden crest]

it probably does considering its novelty

im not saying im completely unskeptical but its not an apples to apples comparison

even a fraction of their benchmarks seems exciting asf

bit curious how runtime works though

I assume it wont use the same gguf plug and play mechanism in its current state

[frozen igloo]

https://youtu.be/4JuHBtquP1I?si=g3Pdblr0z6rXI8fD whoa funky material

And some funky translations

[amber fractal]

edit from the future, it was 152207 times

Amazing levels of overkill

[frozen igloo]

https://tenor.com/view/desert-tumble-weed-ghost-town-gif-11132442

Gen chat is the new programming

[leaden crest]

we must conquer all

tomorrow we take nn

then we do border wars on artist alley

manifest destiny all the way down to bot channel

[flint dome]

you dream too small we plan on conquering other servers too

[flint dome]

hes not wrong the only reason i boot mine is to watch teto sit on my window

[frozen igloo]

https://tenor.com/view/evil-evil-neuro-neuro-sama-vedal987-gif-4551340732191743996

[glass flower]

i should at somepoint try to create something like that but working on linux..

[silent cloak]

We should have a secret programming chat for the real programming

And this one will be for the normies and neurotic neurons

[maiden geyser]

modsssss this person is literally trying to create a secret society

[frozen igloo]

So is this just trained on their own emoji stockpile? Including ones that haven’t been released?

A lot of words just don’t work

[opaque wharf]

Try Neuro Sama

Or Neuro Samantha

[frozen igloo]

Neuro doesn’t work

[opaque wharf]

[frozen igloo]

This is neural Samantha

[frozen igloo]

Oh wow it runs locally

That’s why my phone lags

[flint dome]

sir thats #1336777692070023271

[frozen igloo]

[olive sable]

https://tenor.com/view/gojo-gojo-satoru-hollow-purple-sukuna-mahoraga-gif-13503454303453550028

[frozen igloo]

Wait it’s actually sorta close

[silent cloak]

Microsoft bing game has returned

[frozen igloo]

How small is this model for it to run on a phone?

[frozen igloo]

enlighten me

[silent cloak]

the ui looks like a bing product

[frozen igloo]

O

Wait that’s actually yellowish blue what the

[olive sable]

thats just yellow and blue

[frozen igloo]

Where it mixes

[olive sable]

in terms of subtractive colour mixing, "yellow-ish blue" is green

in addativie colour mixing. yellow and blue is white

[frozen igloo]

What do you see where it mixes?

[olive sable]

green

[frozen igloo]

I see it slightly green but it mostly looks like blueish yellow somehow

[olive sable]

rgb value where it mixes is rgb(183, 213, 183)
making it slightly green

it couldnt figure out whether it wanted to do subtractive or addative colour mixing, so it did a bit of both i guess

[opaque wharf]

It doesn't look green at all to me

[olive sable]

well

this graphics is my job

[opaque wharf]

Well, you're bad at your job then

[olive sable]

well, fuck you then

im litteraly correct

its green

get your eyes checked

the rgb values have spoken

rgb(183, 213, 183)

[opaque wharf]

We have OkLCH for a reason

[olive sable]

i dont give a fuck about OKLCH

there were enough colour spaces already, and yall added even more

[opaque wharf]

But it is the most "correct" color

[olive sable]

thats propoganda

the only correct way to describe colour is quantity of wavelengths

[frozen igloo]

The interface here just looks like vomit

[olive sable]

this one seems to be only subtractive mixing

otherwise we would have yellow

[opaque wharf]

Yeah, and that requires stupid amount of data especially because photons interact with stuff

Which is also why spectral rendering is invented

[olive sable]

but it is the only correct method. everything besides that is cope

for rendering its even more cope. you need to render to the colourspace your monitor uses

[frozen igloo]

“Greenish reddish blueish brown”

[olive sable]

[frozen igloo]

There’s a tiny bit of blue down there lol

A marble that looks like that would be neat tho

[olive sable]

Unlike rgb() or hex (#ca0000), OKLCH is human readable.

i feel like thats just a difference in what you're used to. i can read rgb quite well, but im bad at determining a hue based on a number. but people that use HSL or OKLCH will be able to do so

the brain adapts and whetever

[frozen igloo]

Eww

It did fine for the first few

This made me flinch

[olive sable]

why all this slop?

where does this even come from?

[olive sable]

what the hell

[frozen igloo]

Apple thing

[olive sable]

samsung as one too it seems

Slop

[frozen igloo]

It works without internet and makes my phone lag and get toasty so I think it runs locally

[olive sable]

ye it does

thats the reason only the new phones have it

my old phone didnt have slop image editor and stuff

[rigid snow]

https://cdn.discordapp.com/attachments/1071784467036913664/1460040503138451701/autismu-big.avif

[olive sable]

i guess it makes sense somewhat to use oklch.
but to me its jsut common sense that HSV will be beighter at certain places it you take a nanosecond to think about how HSV even works

the V in HSV is quite litteraly just V = max(R, G, B)

im mixing shit up now

[opaque wharf]

[olive sable]

so if with RGB (255,255,255), you get V = 255
with RGB (255,0,0) you styill get V = 255

[opaque wharf]

No, you see, the issue is non uniformity of the color space. Look at blue on HSV which consist only 10% at most vs oklch

[olive sable]

no, you see, that's just the second issue it fixes which to me is just common sense

[opaque wharf]

It is the main issue it solved. Non linear/non uniform color space

So you can expect 10% increase of brightness in red to be perceptually the same even on blue for example

[silent cloak]

they are trying to make oklch the new standard

[olive sable]

i dont mind that partially

but

thats only applicable to front-end dev

[silent cloak]

it is kind of a pain to migrate to as some converters just suck currently

[olive sable]

you cant expect graphics programmers to start using over RGB

[silent cloak]

well yeah obviously

oklch mostly matters for UI accessibility

especially for themes

[rigid snow]

why not

[silent cloak]

the suffering thats already caused by gamma vs linear is enough

[opaque wharf]

New nVidia 7000 series with native oklch buffer

[silent cloak]

[olive sable]

i wont mind making every game not made by me be 5x slower

[rigid snow]

is it that heavy of a conversion

[olive sable]

every conversion is heavy

[silent cloak]

its a whole other format

u would be surprised how big a thorn in ur side it will be

[rigid snow]

blah blah all i'm hearing is grapfix programer big skill issue

[olive sable]

you have to do it for every pixel of the screen. at 1080p thats 2 million conversions every frame

[silent cloak]

the amount of BS i had to deal with regarding srgb conversions with past jobs

[frozen igloo]

[rigid snow]

you have to do everything for every pixel of the screen. that's how rendering works

[silent cloak]

my GPU is checking every pixel on my screen and it still cant find who asked

[olive sable]

i dont even touch live HSL to RGB conversion, cuz it makes all my fragment shaders take 3 times as long

[rigid snow]

she goes to a different monitor, you won't know her

[olive sable]

no not really. thats not how rendering works

[silent cloak]

its worse when ur giving textures for one type and u have to horribly convert them

[olive sable]

true

[silent cloak]

https://tenor.com/view/cloudy-with-a-chance-of-meatballs-it's-enough-to-make-a-grown-man-cry-die-earl-devereaux-rip-gif-11770651815678506576

[rigid snow]

skil isooe

[olive sable]

you only render x object to pixels if you know it needs to go there

[silent cloak]

my bed has a pointer to ur mom

[olive sable]

doing it everywhere is the skill issue

[silent cloak]

and ill tell u now that its unsafe

[rigid snow]

NAAAH

[glass jetty]

i see no point for hsv outside color picking and searching for area on hsv representation, since you both have lower precision if store hsv in 3 uchar and it takes time to convert

[olive sable]

HSV is for the graphic designers in photoshop

but i eat them if they dont export to a normal png with rgb

[rigid snow]

you will get Adobe RGB (1998) and you'll like it

[silent cloak]

dont even get me started on some of the sprite formats ive received

i had to learn how adobe does all their color types and make my own preset color scheme so I could mass convert

its hard to do without screwing up the art

[rigid snow]

can we just tell everyone who uses ps to ALWAYS SET it to THIS and NEVER touch anything else

[glass jetty]

no

[olive sable]

const multiplyMatrices = (A, B) => {
    return [
        A[0]*B[0] + A[1]*B[1] + A[2]*B[2],
        A[3]*B[0] + A[4]*B[1] + A[5]*B[2],
        A[6]*B[0] + A[7]*B[1] + A[8]*B[2]
    ];
}

const oklch2oklab = ([l, c, h]) => [
    l,
    isNaN(h) ? 0 : c * Math.cos(h * Math.PI / 180),
    isNaN(h) ? 0 : c * Math.sin(h * Math.PI / 180)
]

const srgbLinear2rgb = rgb => rgb.map(c =>
    Math.abs(c) > 0.0031308 ?
        (c < 0 ? -1 : 1) * (1.055 * (Math.abs(c) ** (1 / 2.4)) - 0.055) :
        12.92 * c
)

const oklab2xyz = lab => {
    const LMSg = multiplyMatrices([
        1,  0.3963377773761749,  0.2158037573099136,
        1, -0.1055613458156586, -0.0638541728258133,
        1, -0.0894841775298119, -1.2914855480194092,
    ], lab)
    const LMS = LMSg.map(val => val ** 3)
    return multiplyMatrices([
         1.2268798758459243, -0.5578149944602171,  0.2813910456659647,
        -0.0405757452148008,  1.1122868032803170, -0.0717110580655164,
        -0.0763729366746601, -0.4214933324022432,  1.5869240198367816
    ], LMS)
}
const xyz2rgbLinear = xyz => {
    return multiplyMatrices([
        3.2409699419045226,  -1.537383177570094,   -0.4986107602930034,
       -0.9692436362808796,   1.8759675015077202,   0.04155505740717559,
        0.05563007969699366, -0.20397695888897652,  1.0569715142428786
    ], xyz)
}

const oklch2rgb = lch => 
    srgbLinear2rgb(xyz2rgbLinear(oklab2xyz(oklch2oklab(lch))))

this seems to be how you convert from oklch to rgb.

[rigid snow]

HOLY FUCK

[glass jetty]

just convert, maybe?..

[rigid snow]

i was trolling but now i just have to take all of that back

[olive sable]

nah thats not my job, thats thei job to do it correctly

[olive sable]

this is the simplified one according to the github person. i have no clue if this is actually correct

[glass jetty]

naming....

[olive sable]

i didnt name it

[rigid snow]

also that's js

my favorite shader lang

[olive sable]

lemme see if i can simplify it further actually. cuz you can just simplify predefined matmul into one matrix

and in glsl you dont need a matmul function, cuz its built in

[rigid snow]

well yeah but still it's 3 matmuls

[olive sable]

ill have to open vscode for this, simplifying it in discord isnt going well

[kind nimbus]

had some similar issues with sunshine a while ago where resolutions looked awful when connecting from a machine with lower display resolution.

my solution was to write two bash scripts, one would turn off the connected monitor once moonlight connects and the other script would restore the monitor state again once moonlight disconnects. then i had no issues anymore with blurry resolution on devices with lower resolution screens.

[silent cloak]

id have to check

everyone has a different version for some reason

[opaque wharf]

Or you know, pen and paper

[olive sable]

pen and paper sucks for this ngl

const mat3 Matlab2LMSg = mat3
(
    1,  0.3963377773761749,  0.2158037573099136,
    1, -0.1055613458156586, -0.0638541728258133,
    1, -0.0894841775298119, -1.2914855480194092,
);
const mat3 MatLMS2xyz = mat3
(
     1.2268798758459243, -0.5578149944602171,  0.2813910456659647,
    -0.0405757452148008,  1.1122868032803170, -0.0717110580655164,
    -0.0763729366746601, -0.4214933324022432,  1.5869240198367816
);
const mat3 Matxyz2rgbLinear = mat3
(
     3.2409699419045226,  -1.537383177570094,   -0.4986107602930034,
    -0.9692436362808796,   1.8759675015077202,   0.04155505740717559,
     0.05563007969699366, -0.20397695888897652,  1.0569715142428786
);

vec3 oklch2oklab(vec3 lch)
{
    float a = isinf(lch.z) ? 0.0 : lch.y * cos(radians(lch.z));
    float b = isinf(lch.z) ? 0.0 : lch.y * sin(radians(lch.z));
    return vec3(lch.x, a, b);
}
vec3 oklab2xyz(vec3 oklab)
{
    vec3 LMSg = Matlab2LMSg * oklab;
    vec3 LMS = pow(LMSg, vec3(3.0));
    vec3 xyz = MatLMS2xyz * LMS;
    return xyz;
}
vec3 xyz2rgbLinear(vec3 xyz)
{
    rgbLinear = Matxyz2rgbLinear * xyz;
    return rgbLinear;
}
vec3 rgbLinear2rgb(vec3 rgbLinear)
{
    vec3 b = step(vec3(0.0031308), abs(rgbLinear));
    vec3 linearPart = 12.92 * c;
    vec3 nonlinearPart = sign(c) * (1.055 * pow(abs(c), vec3(1.0 / 2.4)) - 0.055);
    vec3 rgb = mix(linearPart, nonlinearPart, b);
    return rgb;
}

void main()
{
    // lch goes here wrrr
    vec3 lch = vec3(0.0, 0.0, 0.0); 

    vec3 oklab = oklch2oklab(lch);
    vec3 xyz = oklab2xyz(oklab);
    vec3 srgbLinear = xyz2rgbLinear(xyz);
    vec3 rgb = rgbLinear2rgb(srgbLinear);

    gl_FragColor = vec4(rgb, 1.0);
}

now lets see if i can simplify this

this is just that js in glsl

i made a couple mistakes looking back at this code

[silent cloak]

is there unit testing for shaders even?

[olive sable]

if there is i never used it

but i dont do unit tests in general so

[silent cloak]

software dev job or na?

[olive sable]

im in college still

[silent cloak]

knew it

get ready to be made to do unit tests in jobs for dumb shit

[olive sable]

i go of the principle, if it broke i can see it broke cuz its breaking in full 1440p on my screen

[silent cloak]

gets more complicated in larger codebases when u have tons of different people working on different parts

[olive sable]

makes sense

[silent cloak]

thats when u make comments // wtf is this and why does it break blah blah blah????

or u get ones that is just the developer feeling shame for having to use a hack workaround

"// horrible no good very bad hack"

[olive sable]

for our group project in game design. im the only one that there that has coded for more than 6 months. and im also experienced in optimization and stuff.

so im looking at the code from the others with a

[silent cloak]

thats why documentation is important for internal use

and unit testing is just a form of safe guard

it ensures someone else doesnt break ur code

[olive sable]

it snot liek their code is hard to read. its C# and they use decent naming ocnventions

[silent cloak]

C# is one of the easiest languages to read tbh

[olive sable]

but i sometimes wonder if its neccesary to loop through every object to see if we hit the finish

answer is no btw, changed that to a OnCollision() on the finishline object's code

[silent cloak]

kek

Unity my beloved

[olive sable]

i dont like unity ngl

[silent cloak]

https://tenor.com/view/unity-errors-beloved-heart-locket-gif-2040718952292911474

its infinitely better than unreal

but naturally wont be as good as a game or engine from scratch

[olive sable]

i tried to open my unity project from my linux desktop on my laptop

it wante da very specific version, as unity often does

[silent cloak]

thats why i have unity hub on my laptop and desktop

[olive sable]

it opens after 20 min of downloading and unity taking ages to lauch

and boom everything broke

[silent cloak]

weird

my laptop is running mint

never ran into that

[olive sable]

apparently it wanted 6.004, but it actually didnt want that since downgrading the file to an older verison fixed it

well, fixed most of it

it ran, but it wasnt happy about some shaders still

[silent cloak]

weird

sounds like something wasnt in sync properly

[olive sable]

ye idk

[silent cloak]

always backup and make forks (cache and the library folder dont matter)

[olive sable]

fixed some bugs, made it a lot smaller

const mat3 Matlab2LMSg = mat3(
    1.0,  1.0,  1.0,
    0.3963377773761749, -0.1055613458156586, -0.0894841775298119,
    0.2158037573099136, -0.0638541728258133, -1.2914855480194092
);
const mat3 MatLMS2xyz = mat3(
    1.2268798758459243, -0.0405757452148008, -0.0763729366746601,
   -0.5578149944602171,  1.1122868032803170, -0.4214933324022432,
    0.2813910456659647, -0.0717110580655164,  1.5869240198367816
);
const mat3 Matxyz2rgbLinear = mat3(
    3.2409699419045226, -0.9692436362808796, 0.05563007969699366,
   -1.537383177570094,   1.8759675015077202, -0.20397695888897652,
   -0.4986107602930034,  0.04155505740717559,  1.0569715142428786
);
const mat3 MatLMS2rgbLinear = Matxyz2rgbLinear * MatLMS2xyz; //precompute this

vec3 oklch2rgb(vec3 lch)
{
    bool invalidHue = isnan(lch.z);
    float angle = radians(invalidHue ? 0.0 : lch.z);

    float a = invalidHue ? 0.0 : lch.y * cos(angle);
    float b = invalidHue ? 0.0 : lch.y * sin(angle);

    vec3 LMSg = Matlab2LMSg * vec3(lch.x, a, b);
    vec3 LMS = LMSg * LMSg * LMSg;

    vec3 rgbLinear = MatLMS2rgbLinear * LMS;

    vec3 mask = step(vec3(0.0031308), abs(rgbLinear));
    return mix(12.92 * rgbLinear, sign(rgbLinear) * (1.055 * pow(abs(rgbLinear), vec3(1.0 / 2.4)) - 0.055), mask);
}
void main()
{
    // lch goes here wrrr
    vec3 lch = vec3(0.0, 0.0, 0.0); 
    vec3 rgb = oklch2rgb(lch);

    gl_FragColor = vec4(clamp(rgb, 0.0, 1.0), 1.0);
}

actually, could save some cycles at the isnan

[amber fractal]

[olive sable]

[stark needle]

Good morning

[olive sable]

goodmorning

[stark needle]

My eyes hurt

[amber fractal]

Good morning bwaadow

[stark needle]

For reasons

[amber fractal]