#programming
1 messages · Page 484 of 1
Tell it you got cookies.
assuming this is stored in row-major order
why columns instead of rows?
more efficient memory loading, it's called memory coalescing if you feel like looking into it
so nearby threads reading nearby memory can be collapsed into a single physical memory load 
mind you this doesn't apply to cpus
So the thing is.. for most of these effects it's imperative that pixel1 knows what pixel2 is doing. To make cascading rainbow effects etc each pixel must be aware of its neighbours.
BUt
LLMs have no free will so it's ethical to delete them
in theory only for the first image. but then I have to find an algorithm that would let each pixel know its destination and each step there.
definitely doable but might not be worth it to do on a gpu unless the images are quite large
yeah if you have dependencies like that you can't parallelize 
rainbow effect should be possible though, in theory that only needs position and time/step
would have to rewrite that function with those inputs

Would it be better if I just erase their memories of that interaction that made them hate me and then slightly alter their program so they can’t hate me
Let me find something
non-unrolled loops in gpu code are great 
sounds like you need to do a multi-pass approach. and or checkerboard rendering
simple wavefront pattern might do the trick depending on the actual pattern you need to generate
dynamic programming 
so that instead of a gpu thread only accessing and manipulating 1 pixel it changes a area of pixels around it. and then next incantation it changes its location its checking so it never overlaps with other gpu threads. this will half or quater your performance but... its the only way you can do these kind of things
or you go the atomic route but that is likely worse
i wanted to write a gpu only sand simulation so i kinda looked a bit into this kind of topic
(i'm now doing multithreaded cpu sand simulation tho)
Why sand sim on CPU?
its the only way you can do these kind of things
in this case I don't think so
so far it sounds the effects don't actually have any real dependencies between pixels, should be fine with just time and position
because physics, and the sand simulation isn't only moving 1 pixel or something it has velocity. which is closer to a particle simulation than sand simulation
Hold on, I'll make a quick example
they're just implemented that way right now because bad
i mean... yeah if you can do it without accessing anything and just math that would be best 
my keyboards rainbow led is a gpu shader
#define RainbowAngle 210.0
#define BandFrequency 0.5
#define speed 1.0
void mainImage( out vec4 fragColor, in vec2 fragCoord )
{
//Make a gradient from red to green to blue back to red that is moving from left to right but rotated so it moves at a incline in glsl
vec2 uv = fragCoord.xy / iResolution.xy;
vec3 color = vec3(0.0);
float angle = RainbowAngle * 3.14159 / 180.0;
float s = sin(angle);
float c = cos(angle);
vec2 uv2 = vec2(uv.x * c - uv.y * s, uv.x * s + uv.y * c);
uv2.x += iTime * speed;
uv2.y += iTime * speed;
uv2.x *= BandFrequency;
uv2.y *= BandFrequency;
color = texture2D(iChannel0, uv2.xy).rgb;
color = clamp(color, 0.0, 1.0);
fragColor = vec4(color, 1.0);
}
this is running through openrgb on my keyboard 
What is this language?
glsl
ye
(what is glsl and why are method names lowercase)
GL Shader language
Insane saying as a person that mainly do stuff in CPU
This was a .png I made zoom and do freaky colours
How did it know that its prompt was reprocessing, the fuck
Also, in the future (not tonight) I want to include ability to bring a mask that omits pixels based on either hex-value (or name) or lumen level (within tolerance) so I can limit this freaky effect to eyes or shoes or whatever. And hopefully in the future I'll have it detect a designated colour/lumen and if said colour forms a border, anything within border is excluded..
Big step up from E2B huh
Little background task that calls a tool that checks if the processing has finished?
I think I might be memory leaking. Or hacked.. ya'll madafaqas better not be hacking me 😠
(please, I need this PC)
Wouldn't be much point in that would there
Why not?
no, you will become my bitcoin mining node and you will be happy about it >:D
The AI can't actually do anything while the prompts reprocessing anyway
just get more cores so you can treat the CPU like a gpu 
Hm... odd. The frames duration seem to have bugged out.
Intel in 201x agrees with you 
Aight that's it. I'm gonna go scream into a pillow or something. Nini
Use LLVMPIPE to render everything
1000 core RISC-V goes brrrr
You know, even 1000 core is nowhere near modern gpu capabilities
Eh don't modern intel gpus have like idk 200?
Facts
Vega igpus have like 16 iirc no?
200 what? Cores?
Yea
Depends on how you define cores I guess. But definitely not the standard one
Ig igpu are low core gpus ..
double that
Ah tru
There's also the type of cores on GPU
ryzen 5 7600 has
Yea tmus, shader cores, so on
Beautiful

5700g has 8 compute units, 512 shaders-
I need to screeps I'm actually going through withdrawal
I partly feel like I need an in-between room state
because there are rooms I would like to secure and harvest and develop without claiming
Programmers are high on programming 
hi
hi
hi
Proposition towards AI ethics; Would it be possible to ask a model, or indeed the machine to be a model, what it wants to be? Would you get a coherent response back?
lemon
I dunno about that
I'm on no
I don't really know anything about the subject but as far as I'm aware an AI can't have options as it's just regurgitating training data. It will believe and say what you encorage it to say without independance.
with humans there's the whole concept of nature vs nurture, where some part of our psychological temperament comes from our upbringing and some part comes from our genetics or whatever
of course you'd get something coherent, the content will just be useless
but with an AI model, the nurture part is basically nonexistent, everything related to the AI's behaviour is decided at training time
Presumably as of right now it's kinda a requirement that to some degree the machine must have some 'nature' built into it, in our own image, in order for it to form the opinion in the first place
so I would say that the response the AI gives is largely decided by what training data it is fed
and thus not really its own choice
yeah that's where my thought process goes
it's largely decided by its creator
The whole
"You don't ask to be born, but you choose how to live" thing
you can ask an LLM that, and you'd (assuming it's trained as such) get coherent output, and it would be imitation just like everything else LLMs output 
it doesn't really have any bearing on AI ethics I think
but it's not a choice
although there is part of it that's more uncontrollable if you start training an AI with some of it's own data
except the fact that you're a program designed to predict the next word given an input string, then you don't really have a "soul" to live by
then it starts to drift and distinguish itself from the original data
AI responses are decided by their nature and their nature is decided by their designer
so their designer decides what they choose to do
the AI would never provide a response that could not be attributed to the decisions of the designer
@true hemlock
I've provoked philosophical debate in the programmer chat
Holds up an infant human
"Behold, a soulless machine"
this channel is called programming
this channel is called screeps
we dont do baking here
you're lucky NN occupied all of our on topic enforcement
sasalele 
do these combine for a 75% sale or a 100% sale
It doesn't just choose the most likely next word though there's some variation and randomness to it I think so that could be argued that that would be a form of independence
We very much are for the 5% of the time we are on topic

Yes
it's like resistors so it is 25%
50% sale is 50% off
so 25% 
1/(1/a + 1/b)
I surrender the topic back to the people who need it most
Just a thought I was having at work today, didn't mean to start a knife fight in the Parthenon
take it or leave it

How can they miss you when they're dead? 
True!
they're alive
don't scare me like that
you can see them again if you visit sam
Is Sam holding Shiro hostage?
Sam
Shiro
@olive sable
Hugs are a form of holding hostage
holding a hostage directly is dangerous if the hostage is physically stronger than you
And yet cats hold human hostages all the time 

Mario kart Saturday
miao
cats have claws and the intention to use them
so now if I break Sam's structures, he won't be around to replace them
i also have claws do not approach


Played resident evil 1 on the theatre pc earlier as first test
Does leaving this to do its thing in the background make me lazy chat
nice contrast



What's being censored there? Feet?
approaching (jojo)
I dont like what you're implying 
why do you want to know?
The anime girl pfp can't give those away for free 
Its the game, or windows hdr
Yes. From my friend
Detachable feet 
zeppelin speedrun any%

sam might actually be the best way to test my attack AI because I don't have to worry about retaliation
I've found having it on a second monitor like an extra pair of hands kinda nice is the thing
so I can tune the settings
sam send a raid
Attack AI 
my screeps already mark his territory as the next victim so that part works
screeps clone based on cellular automata
Sam is already on his way raiding Netherlands
I just have to enable sending attack groups
A Belgian invading the Dutch 

or maybe evolution sim
Ill kill shiro tomorrow i guess


Another SamOutOfContext material 


skill stats for screeps also be interesting

skill stats?
is that the purpose of the powercreeps?
yee thinking of alterations to screeps
Hey, wait a minute. That ain't nub
ye that bun
trust me you guys seen like 20% of screeps so far 
no, moreso so specialise certain workers into certain fields
i know nothing

at all
Something went wrong with my dual room setup 

Congrats @real sierra you managed to lure Sam to screeps on his mario kart saturday 

screepsday
good thing I did tho, his base looks desolate in that pic
where'd all your screeps go
We're with 5 so loser passes controller. I lost my winning streak
We're watching the greatest showman after last round
Watch interstellar instead 
How does this game work?
You program the bois to do what you need to

luv me lightweight frameworks like cuda

actually how does rocm compare to cuda
whee
assuming custom written gpu code
in terms of
0.0
because I heard about like pytorch having bad rocm support or something
pytorch rocm = you have to install cuda lmfao
power


theoretically mostly the same in the sense that they serve the same purpose
Software support sucks
everything expects cuda so mostly everything is gonna be HIPified cuda
well I write my own
And the HW usually have some gotcha
the hw itself is fine but the hw breadth of support ootb isnt as broad
you don't
by support you mean software being made for it?
to use like anything that wants to use pytorch you gon be HIPifying
Maybe I think the hw is ass because of bad driver then. Been a long time since I use rocm
ye, hipify doesn't need CUDA though AFAIK
Can't wait for zluda
it does
look at the pytorch 7.2 rocm wheel
it operates on source code and just has the function name mappings and stuff hardcoded, there's really no reason for it to
it requires cuda
unless you're using hipify-perl which i've literally never seen anyone do
tthe entire premise of HIPify is that it transforms cuda
hipify is a converter from cuda to rocm no?
generally worse dev tools
yes
it needs cuda
because it's for cuda
lmao
that is the base of what hipify even is
HIPIFY documentation
HIPIFY is a ROCm tool to help developers migrate GPU programming from NVIDIA’s CUDA language to AMD’s HIP C++ programming language for use on AMD GPUs. HIPIFY includes two tools offering different levels of capability:
does that really matter? since I would be writing custom gpu kernels
so shouldn't matter if it's written for cuda or rocm?
if you start rocm/hip native it's fine
ok so we ball
you dont get PTX
Nixpkgs ROCm build of PyTorch 2.11 does not need CUDA
idk why the official wheel includes it
and there's not really any reason for it to
since hipify doesn't need it
why would hipify not need cuda please explain
talking about things like profilers, debuggers etc 
but yeah the code itself is w/e
hipify is for cuda code
has mappings built-in
it doesn't load them from the CUDA toolkit
like

brother it does require it it literally says that in the documentation
I'm probably getting a 7900xtx for my new pc
sounds like it's a better pick than nvidia gpus? at least for the cost
it gets the mappings from cuda
show 
shuni I was confused about this too like a week ago 
surely typo 


the thing is
the mappings definitely are in the hipify code
the problem is
so idk what it needs the toolkit for
cuda has 190345oi82174398749128 versions
you have to target the specific cuda version you need
when you HIPify
so the best way to to do that is just install the pos
this is the overall rocm curse
it has to handle the fact that cuda is the default
and that's just janky and shitty
even if you get something that is like
native rocm
if you ever pull in any dependencies
99% of those will just yoink themselves a full ass cuda installation anyways
and that will overwrite half your shit with cuda versions
and then it's fuckin gover
rocm might as well be named segfault simulator
was that you i had the same conversation with lmfao

looks like the CUDA dependency is for tests only?
why are the tests included in the shipped version and not a separate repo hell nahh
regardless of hipify itself needing cuda to exist in a vacuum
it exists solely to translate cuda to HIP
yeah just never use hipify
and use just raw hip
yeah you can do that jusit never use any dependencies ever
as they will then install cuda
that's been my experience anyways
stuff like triton is pretty much cuda only
for ROCm/HIP, you DONT NEED cuda for pretty much
pure rocm/hip dev
building rocm from source for a specifically amd target (which is what TheRock builds do)
running already hipified code on an amd gpu
hipify-perl
or any of the rocX libs
hipify-clang needs cuda because it actually parses the AST which needs a working cuda install + headers on that system
oh also hipBLASLt for some reason requires cuda lule
it doesnt have a rocm only version and it CAN target nvidia
so it invariably pulls in cuda
it's deprecated tho
This chapter presents how to port the CUDA source code to HIP
hipify-clang is a Clang-based tool that parses code, translates it into an Abstract Syntax Tree, and generates the HIP source. For this, hipify-clang needs to be able to actually compile the code, so the CUDA code needs to be correct, and a CUDA install with all necessary headers must be provided.
hipify-perl uses pattern matching, to translate the CUDA code to HIP. It does not require a working CUDA installation, and can also convert CUDA code, that is not syntactically correct. It is therefore easier to set up and use, but is not as powerful as hipfiy-clang.
ye apparently it's just for the headers, which somewhat makes sense
it's a good option @olive sable will tell you if he's awake (it's almost midnight for him)
the perl version doesnt need it but it will also annihilate your codebase for free without a care
I'm currently dragracing my 5950x vs an aws 64vcpu instance for what can compile linux the fastest, what do we thing is going to win?
still not sure why Nixpkgs doesn't seem to need it, but it also doesn't have many consumers of hipify so mabe that's just an exception or something 
im just guessing for nix that somehow nix shenanigans allows the included shite to be controlled in a way that it magicks that requirement away, or it pulls them in as needed somehow but not initially
or maybe it only uses hipify-perl
idk
since it's only needed for hipify
basically
but yeah I did quite a bit of research on gpus
and that 24gb of vram is just too sweet to pass on
while not costing infinite money like the rtx 90s
that's why i have an r9700
there's a good chance that it does 
that sounds like a nix thing to do
both rccl (why ???) and onnxruntime seem to use hipify-perl
not sure about torchaudio but it's not marked as broken and it uses hipify so it's probably the same
it's not a patch though
not a Nix thing
this is just how these projects are 
that's a nice one
out of my budget for now tho
there's a w7900 with 48gb of vram if you want to spend RTX money
then why is the last commit 17h ago 
tho I technically already exceeded my budget
spending 2 salaries on a new pc and trying to calculate the food money lol
rocm moment
ah it's just the repo that got deprecated
it got shoved into rocm-systems
ye
So is it true that Intel's GPU compute thingy is better than AMD's?
it's cringe there is even a difference between cuda and rocm
can't they all just use the same shit smh
it's not like cuda itself has a shiny compatibility issue free ecosystem in the first place
yeah, it's called OpenCL
not going very well 
intel's is just significantly different

like
most of the ml related stuff for intel gpus are forked
and extremely reliant on intel supporting that fork
what the fork
ye
they went with a very different approach
zero CUDA compatibility, but they actually maintain important parts of the ecosystem themselves
intel uses oneAPI but that's actually like 6 frameworks in a trenchcoat for example
and theyre all their own separate thing from cuda
amd tries to maintain interop but it's a disaster
intel tries to do its own thing
separate disaster for other reasons
pytorch IPEX is upstreamed now i think
which is near
neat*
mechanically, the difference is obvious
on amd/rocm
you can just use torch.cuda.is_available() and it will work
as if you're on cuda
for intel you will need to first import IPEX shit and then monkeypatch it for that to be possible (not sure since it was upstreamed tho)
i read somewhere the intel just dropped their SYCL support
well support as in
development support
in one of their OSS shrinks recently
which is an L
i hope i was wrong tybh that's like most of their stack
oh no it was a FFT lib for SYCL
sheesh
not as bad kek
can't wait for nvidia to drop support for cuda c++ 

sycl is nice but too high level for my taste 
Let’s play ‘how many messages do I need to read before screeps’
screeps talk was a bit ago
now its AI time again
with how many players shiro recruited maybe its time to create a dedicated swarm screeps channel
Screeps goes until shiro gets bored again
maybe... but i do plan to take it on properly now. my old codebase grew in a bad way, was also c#. And also now i know roughly what i need to design my system around already
no more surprise invader jumpscare 
you should play it
That would go crazy
Ai you say?
Guys. just fixed kai's delayed responses. she now responds within milliseconds
it took me forever
my own neuro ive been working on for months
Interesting

How many months?
like 9
complete from scratch
My friend can't english, so I google translated my code
does it even work?
It's a multi-threading VMsDispatcher middle load balancer
(VMs I mean in this context is scripts that doing multiple tasks in parallel)
Anything interesting feature/architecture wise?
Like that I wouldn’t expect
well. mostly ive been working on her personality. she has an evolving personality module that allows her to make her own personality over long chats so that way she isnt a set prompt
but nothing too crazy yet. because shes still like in bare bones. ive just been trying to actually get her to work first. if you know what i mean
@trail smelt please checkdms
i have plans for like game implementation and other stuff. shes not gonna be open sourced because i plan to stream with her. but im not trying to get copyright infringement or whatever from vedal so she is gonna be a completely different personality, model, type of streaming, for example she isnt going to have a singing module as thats neuros whole schtick basically
im trying to make her as unique as possible
Vedal doesn’t own the concept of having an LLM do your entertainment job for you
ok.
i just didnt want people saying. OH thats just neuro. this guy is copying her. etc
i thought. Oh neuro seems so cool. so i decided to build my own. its been such a difficult journey. as im only 18. so i have to juggle with school and all that
No no
They’ll still do that
Trust me
oh i know
Source: I’ve been making one for 15 months
seriously?
hows it going?
are you planning to stream or is it just for fun?
yeah kaine what happened to her lmao
Well mine is focused around being on discord and using experimental tech so the plan changes basically every week
However
I think the big update will be coming soon
ah so not for streaming?
Likely both
Depending on some tech I’m working on
But my expectations for serious popularity are zero
She says she is fine #1333098090772693062 message
you gotta promote. what ive been doing is doing like dev vlogs on tiktok. and people are loving it. i even got a twitch streamer with like 40k subs on tiktok and she commented. saying it was cool
if you do end up streaming is it gonna be vtuber or no
my plan is vtuber. i have her model picked out
Likely but my model was ¢33 on Etsy and I don’t plan on paying $300-$800 for a custom one so we will see
I have no intents of making money with it tho
yeah me neither. im just using default model for now. but not the one neuro had. im using a completely different default model. it was like 5$ a month for access to hundreds of different models
what did they add a sub now?
no. im just using a different program. because the one vedal used originally only had three models.
does ur program have a good api
yes
im gonna implement her model in probably next week or the week after. my birthday is this upcoming week so i wont be able to work on her that day
nice
if i had a live2d model i could put mine on my phone app
it took so long for me to figure out how to get millisecond responses from kai. originally i had to wait anywhere from 3-60+ seconds.
wait i do but its from 2024 and old model and it's trash
in order to get millisecond responses what i had to do was have a fast response server that captured my voice from discord and responded but sent the data to my main brain so it will have all her personalities and such
do you think i could check your ai out. not the code but just like chat with it
so i could get ideas for mine
Uhhhh
Gimme a sec
?
already watched that
I liked the greatest showman. 8/10
pov: pov moment
theatre pc doens thave gpu yet, and eventho res evil 1 would probably run well on the igpu of the 13400 i just stole the 7900xtx from my desktop
apparently gpu swap takes only 25minutes
ye its a good gpu

45W charging is not a joke, jesus christ
form 6 to 70% in 20 min
the ETA of "1 hour" means 20 min for the first 70%,a nd then 40min for the rest???
I'm convinced Capcom employed some sort of dark magic for Resident Evil Requiems graphics
Its ridiculous
This is running at 4K/120fps
Albeit frame gen 2x but even then, 60 is still mind blowing at 4k with a game that looks like this on last gen hardware
AI VTuber singing is like literally not even remotely a Neuro only thing
Also singing is stupid easy to implement
oh. i didnt know that
yeah i know
I specifically specialize in vocal synthesis
interesting. i might need your help later on
Do note you'll need to provide quite a lot of audio data if you want anything to work
yeah fr u just need like 70$ or literally free if u use that oss thing
open-source vocal synthesis is really incredible
All you need is a good GPU to train it on really, and the knowhow of how to configure it
I'm probably gonna make a better GUI for it in Kotlin later after or right before NeuroSynth's release
You know
He actually has this project
He keeps it kinda lowkey so you may not have heard of it
But it’s called Neurosynth
Biggest NeuroSynth fan be like
Is it tho
Voice is one thing
I’ve only had failures with locally running song covers
And that’s without complex stuff like mixing and lyric replacement
Which Neuro has both of
I assume you mean RVC, which alone sucks
Open source machine learning is basically mind-blowing in every respect for what you can achieve on (relatively) modest hardware
From LLMs to vocal synth to VLAs to upscalers
(this is the part where he then mentions some really popular song covering model which I’ve somehow never heard of before)
(and it isn’t Neurosynth cause that’s still confidential state secret)
NeuroSynth's tech stack is able to work just fine as long as you have computer, and it definitely won't randomly break like RVC does
(he failed that second constraint)

This was supposed to be the part where you suddenly recommend a model I’ve never heard of that does song covering really good


If only someone was developing something like that
yep clanker
I'd thought I was decent at the terminal, but I just switched to linux and god damn
I’m in a server called Linus Torvalds fan club
I just got python working again so that's awesome
Use typescript 
not the kind of edgecases i meant
logical error caused by how i programmed it
intended behavior did not translate to actual behavior
Ain't no way lol
beta came out a couple day ago yea
Rewrite it in rust 
nah
rewriting in rust would not fix code flawed at the level of it just being a bad design
Some people would disagree with that statement and insist that rust forces you to write good code because of the compiler
idk how the compielr does shit in rust besides the community not allowing me to do things how i want. but in this case the error was definitely not on the computer side but on the coder side
rust wont fix bad design
Exactly 
Fearless concurrency
look inside
Arc<Mutex<T>>
Sometimes Pin<> too but I still don't know how exactly that works.

What you mean?
not for a long while
shiro marketing pay required
holy necroping
forgot glasses
Speaking of readable code
blame discord 

how is your screeps journey going
alright, although I think I softlocked myself
my screeps are alive for now but they've kinda peaked, they lack the functionality to do much more at this stage
impossible, the spawner passively generates energy
no as in
I am missing some code but I can't view the code anymore
it just pushed me past
so now I can't progress
oh you mean in the tutorial
unless exiting the game restarts the tutorial
just gotta restart ig
:V ok
Wtf is a screeps tho
scripting creeps
Roblox lua devs
Roar by Evil with vocals isolated and reverb removed and then compressed with ~2.5kbps Speex and then converted to a Discord friendly format
here's the original if you want, won't play in Discord
wow, those parts at 2:26 are gnarly
why compress to 2.5kbps? are sharing with dialup users?
idk 
ah, yes, much nicer on the ears...
yes
there's reverb near the end :/
that's fascinating... it seems the harmony eats too much bitrate and makes it turn to shit
huh
it was only 2.5 to 3kbps mhm
Warning: Speex is only optimized for 8, 16 and 32 kHz. It will still work at 11025 Hz but your mileage may vary also saw this.. that's.. interesting
11025 actually sounded the best.. 16khz had this really grainy sound that just hurt the audio quality
cause it makes the vocoder sound
bleh
I also think Speex kind of beat Opus at low bitrates... HOWEVER, I think the Opus speech optimizations starting with 1.5 probably make it way more competitive
but they aren't built by default and I'm too dumb to figure it out, gah
these are the official examples
female_nolace_6k.wav excuse me, I prefer my females with lace
i think my bottleneck is that i only have 1 energy source
took whole night for the controller to almost hit level 3
Speech codec enhancement methods are designed to remove distortions added by speech codecs. While classical methods are very low in complexity and add zero delay, their effectiveness is rather limited. Compared to that, DNN-based methods deliver higher quality but they are typically high in complexity and/or require delay. The recently proposed ...
sureeeely
adatpive
nah, this
so Opus is using a low complexity neural network model with post filtering... I think?
they don't build it by default because it raises the binary size by 1.6mb >.>
I used the windows build but I don't think it had it either, pfft
that's insane, 6kbps NoLACE sounds better than 9kbps without the speech optimization
The LACE model has a complexity of 100 MFLOPS which leads to a run-time cost of ~0.15% CPU usage. The NoLACE model has a complexity of 400 MFLOPS which corresponds to a run-time cost of ~0.75% CPU usage. 
im being robbed 
Sam already told you that you're their buffer 
I find it funny how the quality increases past a certain loss
She coded a calculator in rust and one half of chat is saying "peak content" and the other half is hitting on calculator-chan
QRT: Themyster1ous01
is your oshi autistic?
why is she coding it like that
also no way kali linux in the year of 2026?
looks different from the last time i used it
hi
Toast
linking this and not pinging computed fn is insane
awawawawawa
If anyone is hoarding Linux ISOs here. There is a kings day deal promo today at eweka.
learning rust with vtuber
lets gooo
btw, idk what else, is this kind of complete?
yeah like this
rustuber
I am toast here
ok, starknight#1001

immortal venerable war
hi guys
wrr
D:_ _
immortal venerable wrr is wrr
Wrr is the sound my cats make when they are being pet
He is sleeping
parasocial + rust user = schizo
Use a .png for that instead
you can't favourite a png
Pros of light mode: u don't get trolled
No, but you can put it in the clipboard and WinKey-V when needed.
is it not grey in lightmode?
im losing my mind but i dont want to talk about it
What do yall thinks about reddit
I can't navigate the site.
🍇 🌽 website
Neurosynth
Debatable
you explod

tv

Evil but she's an early UTAU
That is so impressively clean
I need this technology



Why are you cheers
cheers mate

emi 
My gif
wrr
yeah
Why not cheers?
He is purring
car
mods kill this guy
idk, to install arch you have to use live iso anyway
The perks of not using bootmanager is my install never broke
systemd-boot my beloved
i use it whenever i can, it works wonderfully
grub is too jank nowadays in comparison

You don't? It's just one command using efibootmgr
with configuration like everything?
grub use efibootmgr btw, you cant setup grub without this package
Yes? It only requires UUID

I might've actually found a use for a 40g network card
SMPTE 2110 needs 3 gig per video channel
ok
but you cant dualboot without grub, unless you somehow bootload windows from linux
if not more above 1080p
though all of this is just a pain in the ass to actually deal with
you can, any modern UEFI manages multiple boot entries itself, selectable in the boot menu 
You won't have a really cool boot selector menu from GRUB that way though
can't fit 4k on a single 10gbit link
you mean changing boot order from inside uefi gui itself?
not boot order, the boot menu is a separate thing that allows you to boot into a certain OS or device just once without making permanent changes
though it should also be exposed in the boot order 
No, it's white
I can. Most motherboard support selecting boot item by pressing key combination when turning on
Usually you press Del or others to go into UEFI, same thing
When you press F8 during boot it will show the boot menu
let me try
Yeah, it's motherboard dependent
So check your manual if on PC
Btw, what is the difference between percona and mariadb?
not work for me, tried also f9 f10
f2 is enter uefi gui, its all
i can only change boot order to windows and its all
in gui
not in some menu
What's your motherboard?
hibernation is best thing btw
remind me how to ger model from dmidecode
I have no idea lol
Version: X415EA
it's possible that it's disabled or something, as computed_toast said, it's motherboard-dependent
but you almost certainly have it, it's the same menu used to select boot devices

sorry i value going outside over larp
like literally i never used any player app except evilplayer to download new music
for new music i just watch youtube playlists...
idk I use spotify because it has a decent ability to give me more of what I like
and I can't easily download a lot of the music I like
why?
don't have a lot of spare storage on my phone, and I spend a decent amount of time listening on my phone
only about 3gb for me for 25+ hours idk
why load from network if you use it frequently
29h

I also prefer being able to passively find new music
instead of actively searching it out
btw, why mtp sooooo slow
it WAS faster for me to upload from pc and then download on phone
but since everything is blocked now, since i have no vpn implementation for android, i have to use mtp to drop music on phone

Disgusting D:
because MTP is over 20 years old as a protocol and complete ass
built for an era when transfer speeds were measured in kilobytes
I'm not gonna install weird schizo os that turns me into a de-facto sysadmin
install Fedora
Sucks i alr was a rhel user for some time
rhel aint quite the same
Still a ton of stuff i need runs only on ubuntu and not some random systems
what kind of stuff you using that's ubuntu only?
Machine learning, a lot of stuff is ubuntu only
People build stuff assuming ur using Ubuntu
Nvidia dgx os is ubuntu itself
🤔
-> LXC
Docker should work fine in most cases
Sure that also works but i prefer LXC so i will say that lol
but ye if you want stuff to run without containerization or patching Ubuntu is probably the path of least resistance

No thanks im not gonna run nested operating systems

Why does some github profile looks like north korea general

More badges and contribution = make you more human and valuable
(Based on some reddit post)
my measure of worth is based on the fact of how much you can shitpost and occasionally shitpost relevant to the stuff you work on
if your memes are good you are good

Github = human
Gitlab = human
Codeforces = human
Leetcode = human
Hackerrank = human
on that note should I try to utterly torture myself by trying to figure out the proprietary spi/i2c protocol by bitbanging it and then using that to run doom on it
i might do it if I have time
front
@opaque wharf what do you think should I try and waste my time will it be worth it or should I not
is that ripped out of a car?
I think it's fun if you have the time. It'll also teach you how to do reverse engineering of a digital communication
But not bitbang it
Do programmers like GUMI songs?
Kokoronashi when? 
I love gumi v2
The sad thing with SynthV 2 is that the pitch curve can't be manually overwritten, only directed
more of a teto and luka fan but gumi always appreciated 
What's that song?

fire papiyon song
Can I like, have a link?
the electrons yearn to travel with their kin not in the air
どうも、蝶々Pです。
それはね、ここにあるよ。
セルフカバー→https://youtu.be/LfLqRfC4iEY
■Music&Lyric:papiyon
https://www.nicovideo.jp/mylist/7203988
twitter→https://twitter.com/papiyon_cyoucyo
Official Site→http://yu-ichinose.com
■illustration : 24
twitter→https://twitter.com/24unk
http://24phage.web....
No VSQx

Where am I gonna find more random stuff to drop NeuroSynth on
With the power of SynthV pitch models I could technically probably match PB's karaoke pace and make "Neuro karaoke but Neuro has energy in her singing" a real thing
If I really wanted to anyway
If you feel like it you can transcribe it for any vocal synthesizer of your choosing and I would be capable of importing it into NeuroSynth
In the meantime, surely programmers like robots with ears
chat is this it
finally normal ram usage
not somehow using 64gb ram on idle
frigging windows crypto miner

is this idle
I wonder if I can beat screeps before I play it

make gpt 5.5 pro extended reasoning go all-in and cook the ultimate screeps bot algorithm
Well basically, most of the memory is Discord and web memory leaks and nothing that actually should be using a ton of memory
use reinforcement learning
bruh
Like read enough of the docs and the Reddit and shit to make a codebase to autonomously gain control of the server before I actually launch the game and deploy it
i mean that would kinda work with a private server
It's like 30GB web app memory leaks at least
Exactly my thoughts
the main issue with that approach is the progression part. you don't have everything at once. Since you build up throughout the levels.
Switching to Linux made my dad come back with the milk 
bro idk how my speed doubled
the antenna is in literally the same place
connected with the same cable
to the same device
some day i'm gonna run a 15 meter cable till my 10gbps switch
Better driver I guess?
Neurosynth 

i created std::function 
fn main() {
let mut x: i8 = 4;
let function: &'static dyn Fn(&mut i8)->i8 = &|x| { *x = 1; *x + 1 };
assert!(function(&mut x) == 2);
}
what do you mean by that
it is reference constant, right there
?

it wont capture anything if thats what you mean
unless that anything is also static

if your function doesnt live for the duration of the program then your function is very silly
fn(&mut i8) -> i8

ye
i just used dyn object becaus they use dyn object
can fn typed variable be mutated at runtime
ye
This took like.. 4 minutes to do >.<
one is multiple
...no

Don't do that you're gonna crash my incredibly overbuilt high end gaming PC
That GIF lags soo baaaadddd

what i meant is
can be created in multiple places
unlike unnameable function types that are tied to their creation site
Do what?
It's not a .gif
rust rtti
Well whatever it is Discord is not having a good time rendering it
Look chat.. she's happy.
how do i declare lifetime for y here?
Why is the hair so bork'd?=
wait what















