#programming
1 messages · Page 261 of 1
Hey.. so my volume became sentient. and evil. Please help (this isn't a joke)
pc
No like every second the volume is going up.
like mine
i've tried
even with keyboard disconnected?
hmmm
i can dm like video of whats happening
are there volume controls on your headphones?
no
weird
Even the button ui is stuck on my screen since it won't stop
try restart
if still happened try disable the audio driver maybe?
Wait the volume calmed down! but the ui is still stuck on my screen..
ye still jsut a restart is a good idea
yep I will after I finish some updates! Thanks
updates 
Yo
Yo

I HAVE ENOUGH DATA TO TRANSFORM UBUNTU TO AI OS
Now it cuda problem with os
[Hardware (CPU, GPU, NPU)]
↓
[Host OS: Ubuntu core + custom kernel/modules/drivers]
↓
[System services (systemd units) & daemon userspace]
• model-registry
• model-runtime / inference-manager
• device-manager (nvidia/cuda, rocm)
• sandbox engine (containerd/podman + seccomp)
• assistant-agent (desktop/cli)
• update-manager
↓
[App layer]
• Python environments, SDKs
• Desktop apps (Electron/GTK/Tailwind)
• Containerized inference microservices
custom kernel
Will it just replace all permission checks with asking Neuro politely for access?
vibe scheduling
vibe delegating
no idea, v9.0 didn't have it. so it failed at compile
I WILL STICK WITH NO GUI MODE
WELL THIS BLUE PRINT WILL CHANGE DEPENDS OF ME
Or use asm to code !
Maybe I will make own is
Live on Twitch: https://twitch.tv/lowlevellearning
Turns out ARM chips have thumbs! 📚🚀 #Cplusplus #CodingTips #OperatorOverloading #MatrixMultiplication #CodeTricks
🏫 COURSES 🏫 Check out my new courses at https://lowlevel.academy
🙌 SUPPORT THE CHANNEL 🙌 Become a Low Level Associate and support the channel at https://youtube....
x86 and ARM are two of the most widely used Assembly architectures, but what sets them apart? In this video, we'll break down the key differences between x86 and ARM Assembly, including instruction sets, performance, power efficiency and real-world applications. You'll learn how each architecture works, their advantages and limitations and where...
Ai os 
IT WAS PROTOTYPE
Dan it why boot has 58 mb
Option 1 make a huge file for boot which cannot handle (max is 64 or 128 mb )
2 work on rest of
(26 mb bootloader) Other will load from driver
Or usb which iso
Yo wtf?
very real
hardware limitations are very annoying ngl
Vibe compilers for languages
The vibe complier when I define True as false and False as true (it is multi-purpose so it is going to assume it is python or something)
guys I don't think she wants to open the VSCode extension 
She wants some weight
W -= lr * dW
b -= lr * db
if step % 200 == 0:
No rules !
My ai is Alpha stage so no final product and here is change logs
Change : 0.02
ADDING MORE WINDOWS CORRUPT FILES FOR WINDOW USER
Tunning for more realistic output like ure talking to humans
Changelog :0.01
Fixing bigram
Fix weight and
And improving network
Removing emotional tree and Replace with ADMT
( ADVANCE DECISION MAKING TREE )
TUNING THE MODEL
ADDING MEMORY TO AI
ADDING WINDOWS CORRUPT FILES
What are you cooking
idk
I genuinely want to ping heir to ban this account
because it looks very schizo ai
i for idiocy?
(why 0.9, that isn't even out yet??)
did you enable the ws feature when adding the crate? you need it otherwise it won't have websocket support
?
the 100 nonsense ©
what does pmo mean again
pissing me off
you're so mean
ah
AHM
it my script doing not me 🙂
cub
Nice ice cube
ice 
How is everyone
testing your drivers on your main system 
It doesn't work on the virtual one 
but I have a old laptop for such experiments
I'm more worried about how skill issue I am
vm configuration issue
it possible to fool kernel mode ac
why wouldn’t your driver work
skill issue
Im testing ai in Mobile without cutting float
funk cub
colon three
console.error("It's smaller")
🤓

is there a point in buying an 18tb hdd
16tb+ seems to be the sweet spot in price per tb
26TB is the top end, 28TB has significantly higher price per tb
so, yes?
they also sell high capacity external drives for pretty cheap on sale often that you can chuck for even more savings
new hard drives prices are pretty bad right now, try to get one from serverpartdeals or goharddrive or something.. and wait for a sale, I've seen them for around $11/tb
no, used
yeah, serverpartdeals and goharddrive are used
you can def get cheaper, those are the really reliable, very lightly used hard drives
$13/tb might be a lot easier to get, those $11/tb deals didn't last long...
You need buy two though-it would be horrible loose 16tb of unbacked up data.
You can get 16tb drives cheap by taking apart portable drives like the wd book - in sales they can be cheap - doing it years - only once did i get a unit that you could not remove the usb adapter to make it a normal drive.
when did we start speaking in NAS?
is this shit doable or no
its a troll task right?
to filter the weak
you either give up and go on or give up and go away
no win here
no, it's unreadable
Its Lisp?
yes
oh i cut the question itself
i kind of want to solve it but i dont want to spend 3 days solving unsolvable shit

Put it in a compiler and step through it? Be good see how it works. Lisp is a head wrecker with all that recursive stuff.
seems doable? it's already recursive so writing it as a function shouldn't be hard, may have to write it as a piecewise function though
im getting a function but it breaks when i change parameters
it works in some scope yes
Ah yes the ackermann function 
hello mister nex.h
i'll hate myself if i look up the answer and its actually doable
Hello mr davepvz!!
just read the code its not witchcraft
it looked simple at first
i understand the code
it gives nothing
aha nice solution bro
literally doesnt work for (A 1 0)
0 is not a positive integer value 
reading the description do be half the battle sometimes
surely it will not kill my morale 
its just, before there was another exercise that was trick question for no reason
and now i expect others to be same too
also they described this book as weird
the ackermann function is an actual function you can evaluate
it's just uhhh
very very very recursive 
fun fact, they named the aot character after this function

at least i figured out this
maybe im not so bad
i expected it to be some bullshit like
{ something > 0 then a
x = {something > 1 then b
{ etc
i hate when they put troll questions it kills the belief
math 
its like math but they wrap it in programming so it feels bad if you fail
it was definitely solvable i just didnt believe in myself
and in the book
i guess it will have more fun exercises later which i can try to actually complete
also how the hell do one come up with a name for a chess bot
i've been thinking about it from time to time and it seems weirdly hard to do
makes sense since my nickname is literally my name 
too much cheese 
hai gang, dumb question because i like to know (and yes i did google my dumbass question) is "Automate the Boring Stuff with Python" good or na?
Meanwhile my chess bot progress: java.lang.NullPointerException: 'progress' is null
book
I don't know if I'll have one in time
any recommendations if you don't mind? lol,,
i dont read books
are you like new to python or is it specifically to optemize stuff?
new
GET MY FREE SOFTWARE DEVELOPMENT GUIDE👇
https://training.techwithtim.net/free-guide
This python tutorial aims to teach you python as fast as possible. This python speed course will cover all the fundamentals of python and give you a quick overview of all of the main python features.
⭐️ Thanks to Simplilearn for sponsoring this video and...
this goes over it pretty quickly, but im inpatient

my opinion is still one of the best ways to learn right now is to demand a decent llm (any of current Claude, gpt, or Gemini 2.5 or even deepseek) teach you and just interrogate critically. just ask it to explain what you should know step by step and don't have it do things for you
I usually use deepseek to just explain things like documentation or questions like how indexing works in an array in a simple way,
or wonder why my code doesn't run (it's usually indentation</3)
ye as long as youre internalizing the lessons it's useful imo
i am, but I'm the type of person to know stuff but not know how to explain it, if I make sense lol
that's an intermediate step imo eventually you should be building towards verbalizing your own understandings too, but no rush, learn first
damn
i come from 100mbit
over ethernet it's 8gbit
yeah
happy halloween
finally i can download datasets at actually reasonable speeds🥹
you're a fake and a fraud

I need it
multigig gang
I love my 2 gig fiber
Almost never saturate it 
my fiber is technically 10gbit
8.5 down 1.5 up?
but it's only accessible via a single 10gbe port
8gbe down, 8gbe up, the rest is as reserve to ensure availability or something
so it's technically 10gbe but in practice 8gbe
oh. here 10gig means 8.5 down 1.5 up
yeah no here it's full duplex
wtf r u people downloading to feel the difference between gigabit and like 100mbit
datasets of size of like 1tb
we share the 100mbps between 6 people. so barely fast enough for 4K netflix
also fair i ive a pretty low tech life compared w you guys
i legit dont think ive watched anything in real 4k in my life
the speed itslef isnt an issue, its when someone is using all the bandwidth to update fortnite or whatever
ye
do you want a utility to throttle everyone on the network to a certain max so that cant happen
honestly no
fortnite updater of the household located
that just means you'll never use the full bandwidth
it means people can reach full bandwidth using it together but nobody can hog it to the point others get throttled
if implmeneted well
For the 4 of us we are at gigabit, there is some priority systems but not much balencing otherwise
gigabit network would be really good but I'm stuck with 100 megabit
we used to be at 20 megabit, thankfully before all software decided to bloat but we were starting to get 4k tv and stuff and temporarly doubled house capacity.
That was not lasting for 8 people any day of the week
ok i looked up existing methods out of curiosity and Smart Queue Management looks like the common 'best' solution rn
(this is just for my own curiosity at this point)
S
scratch the software bloat part, it was after covid
lol
ok so the only way to achieve this without modifying the router itself is to put it on a raspberry pi or similar between the router and modem
hmmmm

samuel vanuel manuel
hello konii 2.0.1 may I see your changelog

me made the jit more cursed
also now it can speak to the world 
never ask a never ask a man his salary, a woman her age, and a tv her changelog 
where does jit get code to execute? is it compiling it right before its getting used?
with this jit in particular the answer is "sort of"

jit overall seems to work bad for big code then
i also wrote a more cursed version that hands off control to the jitted code rather than calling it as a function
both of them are in there and both of them work
provided you are using gcc-x86_64 and additionally use the compiler flags from build.nix

i think you misunderstand the purpose of JIT
its not typically for compiled languages
its to make interpreted languages faster
by compiling things before use?
ooh so it pays off when we are reusing the compiled things then
without JIT java wouldnt be nearly as fast as it is

tost
hiiii
read the code 


the new version?
well the one you sent above
ye ye seem good
im a bit sad because the manual syscall makes it less portable
"portable"
also the ret instruction
but the problem is:
- i cannot call anything from inside these opcodes (segfault)
- if i change
retforreturngcc reorders the functions and my cursed code no longer works (also segfault)
i can fix the second one but its less elegant
and the first one is "to be determined"
it is the kind of project that would get you fired from a job
are you trying to add a jit compiler to hbvm
i mean that only narrows it down very slightly but maybe thats just because my imagination is powerful
yes
but not a good one
the worst one

why do they not have native jit 
lets see if gemini can guess what kinda project it is from your screenshot
jits are hard to write
also hbvm only has three implementations
i'd imagine you'd need a lot of very good heuristics and statistics for a decent JIT
is hblang very fresh language
oh ok
one of those implementations is written by me
yes, but unrelated to hbvm
go paste the whole code instead, rather than an image
gemini wont be able to tell you anything about what it does with just that image
this is the third stage of my "direct threaded vm" project
tbh it told me what i wanted to know which was just the genre in the first place
what does that even mean
im like kinda new to programming but i learned some random parts of the lore deep and am 100% ignorant on other parts
most lowlevel is still on the 100% ignorant end
ignorance is bliss as they say
google it, its not a rare term should be fine
nice, first result is a dead mozilla blog
GO RUST

is this just each opcode gets its own thread/subroutine or am i misunderstanding
konii i am a rust binary machine if you give me any exact spec of what u want one to do i will get it done [but you have to be exact]
also like micro stuff obv not full systems
1 KiB hello world
// stage 0 - naive vm
switch (pc++) { ... }
// stage 1 - computed goto, basic direct threading (far faster than switch)
goto *optable[*pc++];
// stage 2 - translate binary into series of instruction pointers, jump through the list of instruction pointers
// (about 20% faster by my measurements, pretty good)
goto **program++;
// stage 3 - translate binary into native code via cursed devil magic (no real JIT)
goto *program;
i dont know what the performance characteristics of this latest vm are but i would posit not much faster than stage2
the amount of UB im causing is stunning
now you jinxed it, it'll be way slower than stage 2
** 
truly
yes its computed goto semantics
i dont care enough to explain it

is this not c *
it is
so just pointer to pointer
essentially just foregoing a loop for directly jumping between parts of the code
that sounds like a nightmare to prefetch for
why is switch slower 
you would think so, but actually branch predictors do way better on it
i have some images from my hbvm-c impl showing about 0.02% branch misses
which is insane
damn
because cpu black magic
Multigig equipment gets expensive
I don't even have a proper 2.5 gig switch. I just have a Chinese mini PC with 6 2.5 gig NICs in bridge mode for my router.
switch doesnt use a jump table
at best you might get a cmove (usually faster than a jump, but not in this case, because a jump is still required)
smh dumb switch just get one
i guess it's too unpredictable to construct the table on its own?
i think its more like not all switch code is directly translatable into computed goto
and gcc/clang havent added this optimisation automatically because the semantics are complex
computed goto is more restrictive than switch, it also doesnt make much sense unless you are planning to reuse the table
they always find some weird exception and make it 5 times slower just to get that universality
otherwise you are embedding the whole table in your binary for just one switch essentially
sth sth slow code is better than broken code
computed goto is only helpful for a dispatch loop
which is a niche that is already filled by tail-calling with function pointers
technically
My Gameboy emulator handles opcodes in functions, and tail-calls the next opcode from a jump table. Cleaner than computed goto and trains the indirect branch predictor.
Gets 50k fps on the Tetris start screen.
konii how do i get the benefits of non-atomic vectorised loads when i need to atomically write into all the things i'm reading 
can you fix the magic machine
i have tested tail-call based interpreters and they have always come out about 10% slower than the computed goto version
there are diminishing returns with complexity of course
Weird. Must depend on how much state you need to spill when calling.
except in python's case apparently 
mhm, i expect they did something very wrong 
that and computed goto allows for optimisations that dont exist for tail-call based vms

duffs device or something (no)
oh another reason
when i initially implemented a tail-call based vm, it was about 30% faster than the computed goto version
but, it was because i had made a mistake in the computed goto version
which was pessimising it a lot
computed goto is a lot more temperamental than tail-call
nee
me spend all day making this thing work

ye me vm hypercurse
void translate(flvm *restrict vm, const void *ops[]) {
size_t code_size = vm->code_size;
uint8_t *bytecode = vm->exe->bin;
size_t pc = 0;
while (pc < code_size) {
uint8_t opcode = bytecode[pc];
vm->program[pc].opcode = (void *)ops[opcode];
size_t instr_size = fl_operand_size[opcode] + 1;
for (size_t j = 1; j < instr_size; j++)
vm->program[pc + j].operand = (uintptr_t)bytecode[pc + j];
pc += instr_size;
}
vm->code_size = 0;
}
void flvm_run(flvm *restrict vm) {
#define x(name, opcode, instr_size) [opcode] = &&name,
if (vm->code_size != 0)
translate(vm, (const void *[]){fl_bytecode(x)});
#undef x
fl_program *pc = vm->program;
while (1) {
goto *pc++->opcode;
evc:
fwrite_unlocked("env call\n", 1, 9, stdout);
pc -= 1;
continue;
}
tex:
}
here's a minimal example of stage2 for anyone who was interested
relatively simple
i found it actually beat standard computed goto in every instance though
which was surprising
i expected it to do worse in programs that didnt have jumps
but it was actually faster
not addicted
same but rust crates
wrr
oh god i now noticed why 64 bit atomics didn't work

i was still bit shifting as if it was 32 bit
worth it, surely this isn't just statistical noise 
oop, conversation [filtered]?
Must've discovered a rule. Will avoid that. 
this is 4x32bit vs 2x64 bit purely for querying and it's measuring a lot more tbf
i should do a query only test but 
🐸
ne
nvidia baited me by giving me a ulonglong4 type but it doesn't actually have hardware support 
looks like it is just noise after all

c++ compile times are so fun 

midi right
ye
only the "note on" parts cuz this is for procedural animation
0 0.000 0.154 | 24 7f
0 0.000 0.154 | 2a 7f
0 0.000 0.154 | 2a 7f
0 0.000 0.154 | 24 7f
i have no fuckign clue why its activating these notes twice
i like how you can't even see there are 3 different lines on top of each other 
we love the midi drum mapping for Hand Clap 
does midi player software jsut come with a library of different hand claps?

exponential scaling on the horizontal axis, so the actual runtime complexity of this is less than exponential im gathering

in theory O(1) 
O(n) worst case for standard hashmap, some hashmaps have O(n^2) worst case
same generally true of all hash tables really
technically insert is constant time since it gives up after 500 evictions
not just hashmap
i dont remember the notation for best-case, average, and worst-case time complexity
(it doesn't ever give up anymore after i fixed the hashing)
i can't stand big-o notation
though for things like this where the goal is to deal with massive input sizes it kind of makes sense
i think big-o is actually for worst case time complexity
if i remember correctly big theta is average case and big omega is best case
uuuh

not what i remember
Compiling or linking? I work on a 1300+ file codebase and the biggest annoyance in incremental builds is that the default linkers on both Windows and Linux are terrible. Use mold or LLD if possible.
me not remember well
cuda c++
heavily templated cuda c++ at that too
it takes like 10+ seconds to compile this filter


mapped 
maps everywhere 
map map
map
i have no clue what a Cuica is, but its in the map i stole from google
10 seconds for 1 file is rough. The other thing that helps in my experience is putting common STL headers in a PCH, but I dunno in your situation.
"Open Hi Conga"
@obsidian mantle actually i take this back
i forgot i implemented a version of hbvm in desmos
yee desmos
it have uhh
ye
let me show you one of the functions
here are some of the instruction primitives
ono

its actually pretty simple
to write the standard logic gates that is
the full adder itself is just a normal full adder once you have the logic gates

Oh found the conversation again. Yeah the compiler sometimes just chokes if the gotos get too complicated.
yup, you can mitigate it though
https://git.ablecorp.eu/lily-org/hbvm-c/src/branch/trunk/src/lib/hbvm.c
if youre interested in my original implementation of a computed goto interpreter
i think i want to rewrite it though, there are opportunities for improvements still
yeah idk, i think i'll just accept my fate
(it's not too bad yet and i doubt it'll get much worse tbf)
appreciate it though
oh it's exactly the code that i thought it was
yee
this is not my worst c code
i think that title goes to the "jit" vm

all your code is pretty incomprehensible to me 

One thing I wanted to do, but don't have the patience for is to generate the assembly for all of the opcodes at runtime with xbyak, profile the emulated program, and insert direct branches for the most common instruction sequences to save slots in the indirect branch predictor. Would also help avoid spilling things like the cycle counter. Implemented a couple opcodes but got bored.

seem interesting
im too unproductive to use external libraries
I sure wasn't going to implement my own x86 machine code emitter.
oh i didnt either


doesn't hblang have a custom x86 backend or am i misremembering
__attribute__((aligned(1), noreturn)) void _start() {
uint8_t prog[] = {0, 1}; // program bytecode
void (*exec)() = mmap(0, 1024 * 1024 * 1024, PROT_WRITE | PROT_EXEC, MAP_ANON | MAP_PRIVATE, -1, 0);
void *scratch = exec;
for (size_t i = 0; i < sizeof prog; ++i) {
uint8_t opcode = prog[i];
size_t size = (optable[opcode + 1] - optable[opcode]);
__builtin_memcpy(scratch, optable[opcode], size);
scratch += size;
}
__builtin_memcpy(scratch, epilogue, fin - epilogue);
scratch += fin - epilogue;
size_t stack_size = 1024 * 1024 * 4;
regs[254] = (uint64_t)scratch + sizeof(uint64_t) + stack_size;
// write abc to stdout
regs[1] = 1;
regs[2] = 1;
regs[3] = (uint64_t)"abc";
regs[4] = 3;
goto *exec;
}
my "jit" proof of concept suspends execution of _start and jumps straight into the generated machine code 
the aligned(1) is required for it to function
see this is what i mean with too low level for my small brain
you're allocating like a gigabyte of space, then you do some magic, and then you do some register stuff
... in start? 
yes i deleted main because i didnt need it
main implies that your program has normal control flow
allocate memory -> write program into that buffer -> execute it 
That do be how JITs work
the gigabyte is only virtual memory, its not actually used unless i write that far
"uncommitted memory" it's called

the hard part is always "how do you generate the code"
You see now why browsers have to constantly keep up with vulnerabilities.
They add a step where they download the code from the Internet first. 
i remember when i thought i could get away with using an mmap'd file as a write buffer 
#define _DEFAULT_SOURCE
#include <stdint.h>
#include <sys/mman.h>
#include <unistd.h>
#include <stdlib.h>
static uint64_t regs[256] = {0};
#define ops \
opcode(op0, __asm__(\
"movq (%0), %%rax\n"\
"movq 8(%0), %%rdi\n"\
"movq 16(%0), %%rsi\n"\
"movq 24(%0), %%rdx\n"\
"movq 32(%0), %%r10\n"\
"movq 40(%0), %%r8\n"\
"movq 48(%0), %%r9\n"\
"syscall\n"\
"movq %%rax, (%0)\n"\
: : "r"(regs+1)\
: "rax", "rdi", "rsi", "rdx", "r10", "r8", "r9", "rcx", "r11", "memory"\
)) \
opcode(op1, regs[1] *= 2) \
opcode(epilogue, __asm__("mov $60, %%rax\n" "mov 8(%0), %%rdi\n" "syscall" : : "r"(regs) : "%rax", "%rdi")) \
opcode(fin)
#define opcode(name, ...) \
__attribute__((naked, aligned(1))) void name() { \
__VA_ARGS__; \
__asm__("" ::: "memory", "cc"); \
__builtin_unreachable(); \
}
ops
#undef opcode
#define opcode(name, ...) name,
static void *optable[] = {ops};
#undef opcode
for me, i made c do the code generation for me 
110% of this is undefined behaviour
must be a far away goal then 
right now syscall is my only way of doing IO because calling a function inside these opcodes causes the program to segfault
opcode(epilogue, __asm__("mov $60, %%rax\n" "mov 8(%0), %%rdi\n" "syscall" : : "r"(regs) : "%rax", "%rdi"))
also this exits the program
which is why i dont need main

true
dont need
wouldnt be bloat if the c runtime allowed static linking and inlined main into start
but oh well
musl 
Same to you in the vulkan mines

Saw someone post a ChatGPT hallucination in another server
By naming the bestRecord parameter in the header, the compiler's understanding of the function's signature and argument usage becomes consistent with the implementation, resolving the stack offset difference observed in the assembly.
Sounds legit
https://godbolt.org/z/o6sr9KqKc
template <class Lambda, int = (Lambda{}(), 0)>
constexpr bool is_constexpr(Lambda) {
return true;
}
constexpr bool is_constexpr(...) { return false; }
template <std::meta::info parameter>
consteval bool parameterNameIsConsistent() {
return is_constexpr([] { std::meta::identifier_of(parameter); });
}
template <std::meta:...
uh oh c++26 reflection
Yeah you can share > shortlink
P2996 my beloved 
the difference for me is that konii is also using c, but with functions i didnt even know existed. my c code looks bad in comparison
she uses C premium or something idk
I mean, that C code is also bad, just more advanced 
gulag c more like
hopefully it doesnt obliterate the compiler flags
if it does it wont work
✅
legit can't wait for it to be stable
I compiled the clang-p2996 fork for myself and have managed to crash clangd a few times.
The thought of clangd having to deal with code that uses a lot of reflection scares me
I used it in VS Code and it coped a lot better than I thought it would. Not perfect.
I guess we'll see 
i have a cellar, but no cheese
what is the difference betweene a cellar and a basement?
I'm not sure on how to feel about having this compact mode active
compact mode?
cellar evokes vibes that its completely unfinished, has wine storage in it
like kinda like a cave under your house
lemme just take a picture
basement in my mind evokes vibes of like standard midwestern american finished basements
which are different
and feel more modern
ye, it is just very high density
other people probably have valid conceptions of basements too but i grew up with one in michigan so
I should say discord compact mode, just in case someone didn't get it
I find compact mode is a pain outside of dms
the only thing I disliked off the gate was the time placement, so because I don't really backread on PC I just oped to remove it
though just reducing spacing on discord to the minimum is a pretty good compromise for me
I'd consider this cellar
That looks like pain to accidentally scrape your head on
Yep
I consider basements to just be a fully fleshed room but under ground level, and cellars to be more put together without polish. (Granted I am american, but I don't feel like limiting this explanation to such)
... my head would be smacking into that 24/7 lol
fairrrr
https://fxtwitter.com/arktypeio/status/1983210635266498649 he can't keep getting away with it 
📢Introducing ArkRegex📢
︀︀
︀︀a drop in replacement for new RegExp() with types ⬇️

Seems so
How much did it cost 
300
2ndhand

Helped my sister make the switch to bazzite today after Windows 10 corrupted her shit for the umpteenth time
She lost trust in windows and I said she could upgrade to 11 but suggested trying Bazzite, showed her she can grab all her usual software off flathub and now she loves it
Sam Vi Mile 
It also performs much snappier since her PC is an old one with an SSDHD boot drive and all the I/O ops from that classic windows bloat really bogged it down
fair enough
Hmmm where should I blame on ?
Promising new way to inplement in-parameter continuous learning: https://jessylin.com/2025/10/20/continual-learning/
My network Bro
The other 10% being Fortnite, COD, Madden, Fifa, and most other major multiplayer games
I mean support is actually pretty good there these days
Not perfect, but pretty good
Anyway that's why you dual boot though, best of both worlds
the biggest problem is kernel level anticheat iirc???
and some devs also saying no because frick you
lazy devs smh
frick the devs back
i broke something again 😭
me when I group together a band in space
@olive sable can I ask
Yeah
Mainly that
Active hostility to Linux by devs is the key barrier these days
basicly, if the first primitive in the gltf file doesnt have a texture, none of the files' texture want to load
something is broken witht he loader
Easy anti cheat is literally a checkbox for linux support nowadays 
Okay, the table having a texture map texture
so i jsut manually edited the gltf file's json to make it use material 0
Ah
also, geovanni has a material that doesnt have a texture and jdut says "use colour rgba = {...,...,...,...}, and the engien doesnt support that
so they're just black in opengl, and use the material 0 texture in vulkan cuz i never unloaded it
the reason why im changing this code is cuz i have learned about persistent mapped buffers
this lets me change the matrices in a different thread than the rendered thread, and has less overhead
so all in all very based
this doesn't work in webgl tho 
why is webgl so ass?
web
It's based on GLES 3.0, which is hot garbage because it's designed to be supported on 10 year old phones.
well
i want support for 10 year old phones
but im pretty sure these features should be pretty backwards compatible
you don't 
I've been there before
Mobile GPU drivers are terrible, and you will never convince mobile users that their device is the problem.
OpenGL 4.4 is the minimum for this feature supposedly
which is 12 years old
phones use es apparently, so doesnt work
uh
we'll make the phones use the shity webgl version's back-end, or vulkan
Yeah GLES hasn't been extended since 2015 and it was already out of date then.
reminds me of this, thanks nvidia
khronos group has really given up on opengl i guess
No Vulkan on web. WebGPU technically exists but really only on Chrome.
ye i know
i have 4.6.0 jsut fine
then why is it saying 3.2?
who knows
Do u have something interesting news
weirrd
could be nixos being nixos for all i know
NASCAR suspends Sam Mayer one race for wrecking Jeb Burton after Martinsville event
The version numbers are not the same between GLES and OpenGL
GLES 3.2 is like OpenGL 4 with all of the good features chopped out.
🤷♂️
they could have at least kept the naming good
What ure cuda version ?
What a cool guy am I right guys
miyamoto my lord and saviour
every single succes nintendo has, is due to miyamoto
people don't like the wii, they like the miyamoto
“We are not worthy!”
He seems like a nice guy, he’s not even control of the prices at Nintendo, that’s like the pr or ceo guys
Eyes
i mean, i see it as an apple, but its probably eyes or something
my game history teacher says "they released the miyamoto" like he's some feral animal 
he hasnt made many games recently and it really shows, recent nintendo games have fallen off
oh, something is supposed to be there.. I'm on PC but I've set the background to black 
i accidentaly made the vrchat movement thing
mesa? black mesa? valve? HALF LIFE 3 CONFIRMED?
i wish lol
Could be slow because it's on another drive possibly
that poor drive is begging to be put down 
USB 
It's a portable one, what do you want from me 
-# It's also HDD not SSD
The portable ssds are more expensive tho
And I have a storage problem 
Dell gave a choice with this laptop, you could either get 1TB, or a better GPU
bruh
actually
my mobo has 2 nvme slots
is it easy to move a linux install to another drive?
idk
But I'm not using Linux anyway, I'm using the piece of shit known as windows 11
Although I do use WSL frequently
i curreently have windows on a 1TB nvme, but i switched to linux at the start of july, which is on a 220gb sata ssd
im gonna move the windows nvme drive to the second slot, and ill buy a 2TB nvme for my linux i think
My laptop is this: a Dell G15 5535
There was the option to use Ubuntu instead of Windows, which I should have picked, but I like to stick with what I know
Do you know how to make a shader to make an object glow btw, I'm not great with Unity
@glad path Do you know how to make a shader to make an object glow in Unity? neuroDinkDonk
i guess dot product of the view vector and the normals of the object to get the edge, and colouring that? 
I don't know anything about shaders 
well, uh
idk how unity shaders work so me neither
i only do opengl and vulkan
which use glsl
Unity has their .shader language which I assume is different
HLSL cant be that different from glsl 
smh prizmotic not answering evilSMH
you shoudl see a normal in the vertex shader?
Hold on, I just need to wait for a bit
bruh
never used unity

would look like this
For some reason, you can find every possible thing you can imagine answered for 3D Unity projects, but for 2D projects, you're just on your own
called a "fresnel shader" apparently
i guess its not really glowing around the object, only on the object itself
yeah
i guess you'd want bloom?
yes
is that not built into unity?
idfk
I think I found something? But I'll see if it works
actually i found a video
🌍 Bundle MASSIVE Discount 98% OFF! https://cmonkey.co/humblebundle
✅ Get FREE Assets https://assetstore.unity.com/?on_sale=true&aid=1101l96nj&pubref=unityglow
🌍 Get my Complete Courses! ✅ https://unitycodemonkey.com/courses
👍 Learn to make awesome games step-by-step from start to finish.
🎮 Get my Steam Games https://unitycodemonk...
I think I'm going to use VSCode instead of Visual Studio, I don't have the RAM to keep them open at the same time 
why were you using visual studio to begin with? 
It has a workload made specifically for Unity development, I typically use Rider, but I don't have the storage to reinstall that rn
brb
Actually, I wonder how good fleet would be with this
It has gotten better, also it wants me to use a github access token to open the unity project
What I like about fleet is instead of it being essentially the same no matter what you have open like vscode, it can change the layout to suit the type of development
I think vscode can do that too now finally with profiles
Never felt the need personally
race conditions are driving me insane
I love xkcd
idk
polyglot package manager with universal api/adapter layers for every language + upfront forced reporting of deps costs
I do love claude sometimes
lol
Oh, this one is great too
nah, that's based
If you care about what a password contains then you're not storing them correctly
i mean also true idk what kinda password processing these days wouldnt be robust to that
but xkcds from the 2000s those were diff times
I can't remember when this one is from, I just have a whole folder of my favorite ones
@hearty notch
The title of the image
lmao
smh no password length limit 
lmao gelsinger is back on his shenanigans again
templeos 2 trust
Why limit yourself to valid codepoints?
submit arbitrary bytes when creating your account to crash any servers in Rust that try to use plaintext passwords

It's my personal favorite
om
Well, I somehow managed to get the lava lamp to work in home assistant.
It's at 100% battery 
Ok, controversial opinion, but TypeScript is better then python and easier
why does my code have so many race conditions


!bwaa

i have an idea

I cant stop thinking of clicking your nickname because it looks like a spoiler 
same

You two are the second and third to say this 

i need a local file compressor does anyoone know where i can find one?
winrar
Hi listerine 
7zip
listerine? 
???
this u?
I already made that smh

One of these mhm
Windows added support for a few different format in a Windows 11 update.
.tar.zst is usually the best compression
wait fr?
Okay Lysterine
mayhaps....
its lysara
oh
ffmpeg voodoo
mp4 video compressed
does that work when you try to view it?
i didn't fix my race conditions, i only moved them

Look into handbrake
well ffmpeg can generate a video in whatever codec you want
problem is figuring out the command line options
handbrake probably is easier. never tried it
vlc media player can export files too 
It's very intuitive in comparison
Well anything is intuitive compared to ffmpeg I guess
Handbrake is okay, the defaults may not be great, I suggest you look at things
and it's forced 4:2:0 subsampling, won't let you change that
yeeeeah
basically all video is 4:2:0 subsampled

so i have to do that on the terminal?
You have to install ffmpeg and then yes
surely making all 5 of these atomic should work 
x264 and AAC in the year 2025 
param(
[string]$inputPath,
[string]$outputPath = $null,
[int]$preset = 4,
[int]$crf = 40,
[int]$keyFramePeriod = 150
)
if (!$outputPath) {
$outputPath = Read-Host "Output path"
}
$vfparams = "zscale=t=linear:npl=100,format=gbrpf32le,zscale=p=bt709,tonemap=tonemap=hable:desat=0,zscale=t=bt709:m=bt709:r=tv,format=yuv420p,scale=-1:720"
ffmpeg -i "$inputPath" -c:v libsvtav1 -preset $preset -crf $crf -g $keyFramePeriod -vf $vfparams -sws_flags area -c:a copy $outputPath
my incantation for recompressing HDR screen recordings for discord.
That works too
Oh right, I should be using mpeg1video
Discord login test
Main site to external site transfer data experiment
im actually glad that this worked lol
That sounds so cursed 
I don't know if I should be doing this
@lament igloo
now check is_lock_free
I'd expect it to lock cause there are no atomics big enough
@lament igloo tf did that even do
it only needs your username and avatar
You're capped at 64 bit atomics
Refresh the page
bwaa
Well hardware atomics
Does this even make sense? I assume you want all of these to be updated at once.
You really should just slap a mutex on it or use some kind of queue to send it
Pog
the problem is this fucker right here
Just lock that section

Well whichever part you need to
I should probably do a setInterval to check if there's any data. What do you think
Sure
LMFAOO
the player and stuff gets updated in thread 1, while the rendering happens in thread 2
ideally id have just the view matrix and those 3 dot products atomic but alas everything i try fails 
Why are you loading the same one 3 times 
Thankfully I have a chrome extension to ignore debugger calls
im still writing this
i copy pasted that
Ah okay
I got the "Cringe Cheater" achievement 
Gimme the extension pls 
Also i put that there cus u can actually get an achievement when you open devtools
Well, you can cheat achievements 
AAC is fine
ffmpeg AAC, is not
I'm too tired to think of stuff so good luck
Heres the extension https://chromewebstore.google.com/detail/anti-anti-debug/mnmnmcmdkigakhlfkcdimghndnmomfeo
nah, i need to be awake to go after the 3d printer im buying
eh, yeah isn't Apple's AAC encoder the only good one?
can't you get FDK AAC working in ffmpeg? at least that'd be a decent AAC alternative, you can also just use Opus, though
would rather just use Opus
actually if i just dont sleep too long it shoudl be fine, gotta make sure not to oversleep tho
I don't like how noisy Opus is.. it also seems to have issues hitting transparency
@lament igloo Note that you should set it to only be enabled when clicked on, otherwise cloudflare captchas will have errors for some reason
i had issues with opus with vlc, but only when not using the windows store version of vlc for some reason
Opus is also really horrible for stuff like piano.. it loses to even MP3 big time
Probably had an old af version of vlc or something
the only reason i use it is for 5.1 surround sometimes
Or they patch the ms store one idk
Thanks
Why does this guy have 5 YouTube accounts 
Would you like to find out
Read the other message too, or you will have a bad time with cloudflare
Ye im aware 
Why is it so big 

Hi again lysterine
its lysara
🙏 one day i will make tools free such as https://neuro.nya.pub/tools/prefixer
Okay Lizard
ADD A FUCKING DARK MODE, IT'S 1:00AM HOLY FUCKING SHIT
nah it jsut doesnt run at all.
ld.lld: error: undefined symbol: __atomic_load
>>> referenced by atomic:308 (/usr/bin/../lib64/gcc/x86_64-pc-linux-gnu/15.2.1/../../../../include/c++/15.2.1/atomic:308)
>>> output/debug/main.o:(std::atomic<glm::vec<3, float, (glm::qualifier)0>>::load(std::memory_order) const)
...
...
LMAOOOOO
Alrllllllrrt

Why does pressing the login button again log you out 
and then notice how horribly slow it is and go back

im probably gonan try to do double buffering hopefullty maybe it works
onyl an atomic index should help with how shit this code looks rn
If you just need to wait for something to be updated before doing something else there are also condition variables but idk if that helps here
I tried doing
btn.innerHTML
And it didnt work lmao
its not really i want it to wait, it's more like, use the new data if its there, otherwise use the old data since at 16K fps you wont notice anyways
Why does removing /discordtest/ from your link bring me to "neurows XP" 
Why are you going to /test 😭
Theres nothing there
wdym there's Neurows XP there 
Thats the fallback page lol
just for my own sanity, you're hyperblaze right?








