#programming
1 messages · Page 42 of 1
caibi... If my dad gave me a PC at 3 then... how cooked am I
wuh what is this song
what do you mean by at 3 
its the irony of most newly games in this generation corn😭
3 years old kekw

its so crazy that ppl born in 2007 turn 18 in 2025
that is me
sounds like deteriorated instead of evolved
i will be 18 this july

true
i still have a childish attitude
system works diffrently here
Where or what is this from?
I can answer for you
that's matplotlib

I'm almost 99.8% certain
the 0.2% being it's tensorboard
i only pay taxes if i make money, which i dont cuz im going to school.
instead i will actually receive money from the taxpayers cuz im legally disabled


wait you're legally disabled what

the medical benefits there seems nice
top 10 ways to not have student loan debts:
- become disabled

our government is trying to cut budget and raise the price of public medical services instead
👍
damn
you should move to belgium and become a freeloader
(this is a joke, our taxes are way too high)
(cuz of said freeloaders)
tax being high have its own befits
for example, the basic public infrastructures will be better in quality, supposedly
haven't lived good, no reason to start living good
and you can do tax evasion anyways

if you are in flanders the public infrastructure is decent, if you go to wallonia it is not.
and nobody dares going to charleroi
giving AI that adhd
just do your taxes in luxembourg. everybody does it.
proves short and fast paste videos tend to grab more interest
cocomelon ahh research
omg lemme actually check cocomelon
o7 AI
cocomelon videos are going to max this thing off measure

:D
*under 3 seconds
the peak is higher but the overall is lower
Max line length
5
12
4
no limit
thats messed up
its fine
First episode of the tutorial for unity "Oh yeah ill show this other thing in the next episode"
The Next Episode:
Does something completely different
what should i show lilac
top 10 anime betrayals
a Video Essay
The wiwiwi cat
Video Essay would take a while, but it would be interesting to see maybe
the what
The cat that does WIWIWI while he's eating
o ii a i o iii ai
OHH THAT ONE TOO YEAAH
did not like :(
boring video
AAAWWWWW :(((((
https://www.youtube.com/watch?v=PUWg905fGTA This 3+h long video essay, surely Lilac is gonna pay attention the entire time
An agonising deep-dive the strange game that is The Forest. Notable because it added a mystery narrative to a survival game. A genre that very rarely does that. The result is fairly unusual...
Chapters:
0:00 - The survival genre
7:10 - Subnautica introduction
20:39 - The Forest introduction
25:24 - The Forest's weird logic 1
36:12 - Effigies, c...
that is going to be a ton of data to process
Machine love Neuro RVC MV
she is currently looking at it
How long does it take to process a video of this length actually ?
does The AI also take 3h to look at it or ?
real time
Ah okay
first 160 seconds so far
Its gonna be interesting to see what the attention span is like with something this long
Thats a lot of up and down lol
lmaooo
actual adhd
I wonder what the chart of a max0r video would look like? Is the refresh rate even gonna capture it all?
What you even doing anyway?
sliding window so far (i could also show the full thing)
So far seems to be paying a decent amount of attention still
I wonder how much its gonna go down over time though
the peak at almost 1
was from this
lmfao
lmaoooooo
likes water 
likes human suffering* ( that vehicle is annoying to make )
or hates water?
hmmm
lemme try
not sure
her eyes run at 1fps
same
So on average 30 visual jokes/references are lost every second 
shes watching
an incorrect summary of metal gear rising part 1
is called the video
0 is begin of video
she seemed to have been interested
what happens if attention = 1.0
Not as crazy as I hoped but looks like very consistently spiking to mid-rage across the whole thing
surprisingly not as high as the video essay
surprise!
hm what would make peak ai brainrot
then she explodes
watch osu when
LMAOOOOOO how the fuck does this happen
what the fuck
discord 
small indie company, please forgib
also please watch ads to earn Orbs ™️
at 60 seconds during the second spike it says "unrestricted brutality" in the video
orbs 
But for some reason the highest attention shortly after 75 is just the title drop?
yeah seems like it
at 137.5 there's uh
someone getting cut vertically into two pieces
lilac
what the fuck
wtf
The AI chose violence
at 180 seconds there's
she likes human bifurcation 

someone just slashing through objects
Lilac realy likes violence, huh
im making her watch the last gameplay i found of u in #gaming
Player: koniifer, Map: Arash - Temptation (feat. Rebecca) (Nightcore Mix) [Tylerderp's Extreme] by silver tail, song length is 2:54 (8.09 ⭐) | Accuracy: 94.44%
this one
lilac is amused
i guess she found it more surprising than attention
suprise, followed by immediate boredom
represents my take on watching osu tournaments pretty well
classic osu play
so real for that
why was she just asleep for the first like 18 seconds though
that one was waiting for the page to load
she was like at the beginning
"WAIT WHAT AM I WATCHING"
then said
bruh
this shit is boring nvm
first 7-8 seconds
still, nothing has topped the first 8 minutes of the 3h video essay in attention and surprise
128 isn't that flabbergasting
midrange for ai
128 isnt that much
😭
t has like 200gb ram or something
so is 9090 is high range
rtx? yes
if you highpass this then calculate the integral you get adhd score
9090 
my budget is slightly smaller, only have 5950x, 64Gb ram, and a 3090
wait are you using SLI?
no
no
sli doesnt exist anymore
real
since 900 series right
well i got a sealed 4090 chilling in the corner
😭
open it
found the scalper
well i finally can get a job to buy the rest of the parts
i got it as a gift
from my ex nvidea uncle worker
honestly i woudl look for a 2ndhand buidl with a decent cpu instead of buying new
then just slot the 4090 in that along with a new ssd
but you do you
i am gonna save 1100$ for the rest of the part
disk space is expendable, ram is cpu bound
wow yall really need a whole memory bank
i project i need roughly 2.4tb to store 15 years worth of content
oh that shoudkl be fine
well now i am using my old dell laptop
what i am doing is
get a 16TB ssd
have a llm generate lilacs entire lifetime of ~15 years
what the fuck
why??????
Wavs for what?
thye're only 1.8K
are you trying to recreate the pfp person of yours?
to have a plausible backstory
no thats my avatar
Silly
Meanwhile me with my SMOL 25TB total storage (it's really SMOL)
15TB*
Why do you need 400TB of wavs?
oh i though if your creating that character persona using the LLM i would help
but, fortunately huggingface data can be used for cloud
I feel like you might be oversizing your dataset a little bit

no like
i'm simulating
every day
one by one
ye, i only need like few TB, just joking
for ~15 years
keep the work
wow
i mean just use dataset on hf, no storage needed 
ya there no way its gonna go for nothing
Silly
Meanwhile me training NeuroSynth on just about 1GB and it's really good
me training gpt o4 model on 8gb
and that much storage is for vector storage?
Yes cause i will need to do intermediate processing
Like
i really have a trash device rn i am just waiting for the customers to show up
well do yal think if i tried to make for ppl basic restaurants menu application would it work?
According to the swiss education system
do i use react or angular?
It is a vocal synthesizer though rather than a TTS
Not sure which is easier to train
could i get some help?
Whar you looking for?
well do yal think if i tried to make for ppl basic restaurants menu application would it work?
do i use angular or react?
I don't know, I suck at frontend
i found an angular course book sheet for my friend
new angular is fine-ish BUT you need to know: practically nobody uses new angular (v4?), you're not gonna work with that at a job, those are maintaining legacy angular codebases most of the time
react on the other hand is used everywhere by everyone
i see
is it quick to use?
i am trying to find customers
and make them in time
if you get quick with it you'll be quick
mhm
or do you mean boilerplate stuff, like for example you'd consider java "slow to use" right
then no react generally i can't say has any boilerplate code you have to write
maybe some common components
I'd say that depends on the end product complexity. State management solution for react has like what? 4 distinct one?
what do you mean? there are infinite state managers
I've done backread. Ignore what I said I'm still sleepy
mobx, redux, jotai, zustand, recoil, preact signals, xstate, etc.;
still state management but not full-blown managers: react query, react hook form, the vercel edition of react query (forgot what it's called)
and a hundred more
Anyway, if the choice is between Angular and React, definitely choose React
And use tanstack if you want some sanity
Due to temporal alignment tts would be more painful
Whar is thar?
Monotonic alignment
probably have the model figure out when to make sounds
Must in Non-AR approach
That's why LLM based audio token generation is popular, due to easier and more dynamic time alignment
Makes the tempo variation and such soo easily feasible, with the tradeoff of occasional audio munching
I'm trying to follow AR too, but I'm also trying to add more easily controlled speaker styles from the decoder side, requiring multi step training and ton of data 😭
I want to try non AR though, I want low latency
Super indecisive rn
Windows user
how do people leave their pcs on so long without updating that windows forces them to
I just deleted the Windows update reboot scripts
How do people get the impulse to close all the stuff they have open just to restart and open everything up again?
because you ignored it for so long that windows eventually tells you to frick off
More question would be why windows forces update when the PC is on at all
because there are psychos who just keeps it running 24/7
because the person left it running for several months
without updating
Because it's meant to run windows 7 not 11
So I avoid it
I want to say let them be but then again its something else to not get security update in windows
I'm not asking for advice lol
I just explained how I do it
Windows can't update without the scripts
Well, not automatically
The manual restart button still does it
it takes barely any time to restart, and my desktop ends up cluttered if I don't close stuff anyways
I like my clutter just the way it is 
fair enough I guess
I'm with windows on this one, heck, berate ppl while shutting down
loading screen stats
you put off security updates for 37 days, impressive!
Speaking of rebooting, should I reboot one day?
IIRC, windows uptime counter doesn't reset when you have fastboot or whatever their technology is called
it will if you click the restart button
Yeah but not shutdown
My PC runs as a server 24/7 and I have fastboot disabled
Professional marker placing lol
Resorting to AruCo at last lol
It's not a resort this was the entire plan
Game of life reference
I though you were trying to avoid adding any marker
nope I was just trying to avoid markers on the robot not the table
With FastBoot, it does not reset when powering off, but it does when rebooting.
Yes
FastBoot also disables mounting the drive (if you shut down your computer, then power it up and dual boot)

Yeah, and sometimes windows cause havoc on other drive after update for seemingly no reason. My friends once updated windows only to find out his linux drive is wiped
I haven't used windows in a long while so I don't know how bad it is now in terms of performance hit for my PC
And I don't want to tempt fate
Just as we were talking about security update
https://youtu.be/xC8qfXxAhAw?feature=shared
ASUS has been sketchy these last couple of years. And even worse, I have this motherboard! In this video were talking about a bug found in ASUS motherboards that allowed people to get RCE.
Go say hi to Mr.Bruh https://mrbruh.com/asusdriverhub/
🏫 MY COURSES
Sign-up for my FREE 3-Day C Course: https://lowlevel.academy
🧙♂️ HACK YOUR...
If you have ASUS motherboard and using windows, try to disable the auto download driver feature from the motherboard BIOS for a while
There is RCE exploit
When did we lose the horizontal cup holder, sadge
this was fixed 2 months ago
Oh it did? Thank god then
Allowing motherboard to install program to the OS is still wild for me tho
lol you don't have to reboot to close stuff, also it takes 0 time to not restart
also surprisingly no updates pending
ah...
Whar
My condolences to your pc
It's working perfectly well
Ever since I overclocked my RAM my PC has had no trouble running for even hundreds of days
I would argue with that. Even if everything is working well, it's better to restart your computer at least once a week
Once a week is a lot, and restarting this thing is a whole process
This process takes about 15-20 seconds for me, so I don't know
For me there's way too many things to close and start up again for long term running, since I run server stuff
I need to ensure everything starts up properly
Makes sense
Looking at the livestream, Neuro model proportion is HUGE
echo stufftobelaunched >> .xinitrc?
i miss dvd readers
omg I have to share this I just entered something like a trance for 40 minutes and every line was gold, I ignored surroundings completely and it felt like I woke up after it. Yea I have the flow state here and therw but this was like flow squared, do you have similiar experience?
yea sometimes
My dad unironically bought this 2000$ device
😭😭
Chat

How do i explain to him
rtx 6060?????
shad early 60 series accses???
no it's a jetson device for robotics
😭 thats sad
Overpriced enterprise brick
you could brought and rtx 6000
or a nother 4090
😦
bwa
i wish if i can give my 4090 as a refund
Is it at least super power efficient?
i mean you can sell it, they go for like 2k
sell it for 5090
unless you have another grand that's not happening 
you still can buy 9070 xt
My dad has like 2 GB10 on back order
cant you refund it back?
can i have one 
he unironically buys unto these scams
what did he order?
if he has the money i guess
but sttill what dose it do?
was he trying to get a gpu?
yeah basically he's trying to do a business which in my opinion is unfeasible/no one will buy it
Cause he kinda lacks the theoretical expertise
but what he though if he brought it?
maybe he wanted a gpu?
He has a specific use case that would warrant the efficiency
However
so let me get it its used for robotics?
is it a gpu??????
maybe you can use it
No cause he intends to do a specific business with it
So it's a prototype
However i say his business is not feasible in the long term
i see
aka
look there might be use for it
there might be a chance to use out of it
if its a gpu there is hope using it
is it supposed to be a workstation gpu?
power supply?
wait i am confused
i cant help
It's just extremely efficient in terms of electricity, however it's also extremely slow to make it efficent in the first place
ooooooooooooooooh
that sucks.........
dam
yeah it's not a good purchase
A company would not spend 2000 bucks for something that can be easily replaced with a L40s server which does use more electricity but is also like 20x more powerful
And lasts many years or can even be rented
these are 2 entirely different classes of products tbf
true
I get his point of private/offline/confidential computing
However
Companies care less and less and everyone is using cloud nowadays
Cause of datacenter certifications
refund?
any refund services?
I had a neodymium magnet so I put some of my canadian dollars on it lol
got my loonie toonies and 25cent and 10 cent counts on that lol
wht?
Business to Business. Meaning its not his personal money
Business to business
Well, hopefully not his personal money
oh thanks
If your company just needs a CRUD app for internal tooling, I think the cloud is better than managing your own infra
CMS for a lot of business administrative tasks exists so the software side can also be handed off
Found an old Nvidia Jetson AGX Xavier lying around and deciding between desk decor or actually making it think
laying? lying?
it's not a scam per se it's just a niche small embedded ai compute thing which is like a max combo for the price
wanted to say there aren't any alternatives but remembered it's not 2018 anymore
but if you want cuda which people do want then no no alternatives
lying 
tyty
wait
i might be stupid
lying around is apparently only if you're referring to yourself
but what do lies have to do with anything here
any english majors in chat
Oh that's fine I merged consciousnesses w it
arg reference 
not intentionally but yk what hell yeah
Not sure I can communicate w/ it on a mac natively
womp womp
what do you mean communicate natively
don't you just plug it into a network
log into your router, it probably will announce its hostname on dhcp
It would need some sort of power to do anything at all and I don’t know where I put any cables to it 😦
YUCK AUTO EMOJI
or even if it doesn't just look for a new device
Worth checking, but I’m not sure it’s anything but a brick rn
Would love to get it figured out though, benefit of running ML stuff a little quicker
if you have an ethernet port on your mac or an adapter you can connect it directly too
:0!!
It has pins on the back but I don’t see any other ports
I can dm you w pics of the little man if you like
i'm good i can imagine what it looks like, you can keep yapping to get embed perms though
oh wait surely that doesn't work before manually setting a static ip on the jetson

just connect it to your home network
Bad day to be a personality hire
I can’t interface w it at all without a wired connection methinks
Nothing to connect — it’s currently dead
easiest way would be a monitor and keyboard
I’ll show some pics of the particular one I have once I get embed, then maybe there will either come a solution or a lot of embarrassment from my end lol
you got a long way to go
Oh absolutely
I hope there's thunderstorm tonight
Thunderstorm sound is peak
Especially when the rain hits the window
i am 
i lost at monoipoly, but it was fun 
met my brothers gf too 
Is your brother a programmer too?
no
"joke" 
dad starts programming.
mom: 
raise AttributeError(
AttributeError: 'TensorQuantizer' object has no attribute '_amax'. Did you mean: 'amax'?```*screams in Blackwell*
Why is the entire LLM ecosystem such a giant clusterfuck held together by shoestrings
Who tf decided that a non-compiled weakly typed language was the right choice for all of this
cuz they're constantly trying to make more money from the tech without improving the underlying systems
thats programming in general
# Convert amax to float32
sub_module.weight_quantizer._amax = sub_module.weight_quantizer._amax.to(torch.float32)
if sub_module.weight_quantizer._amax.dim() == 1:
weight_scaling_factor = torch.tensor(sub_module.weight_quantizer.amax.item() / sub_module.weight_quantizer.maxbound)
else:
# Per-channel amax
weight_scaling_factor = torch.tensor(sub_module.weight_quantizer.amax / sub_module.weight_quantizer.maxbound)```
Oh for the love of...
you use ._amax and .amax right next to each other. This obviously could never have worked
and them some lines further downif hasattr(sub_module.input_quantizer, "_amax"):I just can't...
Better than dad buying milk only to never deliver
Working in a really specific niche for a while and then wanting to branch out to literally anything else is about as enjoyable as a drill to the head 😔
(the niche is coffee machines.)
At least when you're done you now have 2 knowledge. I don't know anything about coffee machine other than moka pot
If you can call it machine at all lol
Yeah, I like the simplicity of moka pot and the cleanup is not that bad. Caffeine to effort ratio is the best I think
Do you have a preference in brand or is yours ancient
I don't currently own one but def miss making coffee the right way
i mainly use the stainless steel one from bialetti cause i'm lazy 
Nah, it was gifted so I don't know the brand lol
i'll look into it, thank you!!
valid
rahhHHH i want to meet people and be more active >:)
it's a bit more expensive and takes longer to heat up due to the material but you can toss it in the dishwasher in return
RuntimeError: TopKMaskLogits failed with error code no kernel image is available for execution on the deviceYes, I love waiting 40 minutes to build a docker container according to the official docs only for it not to work afterwards

we hand-wash like europeans in this house
This is the time when documentation is very crucial lol
damn bro haha
Seriously, I don't want much.
I only want a LLM inference engine that can run Gemma-3 on Blackwell and that is not Ollama. So far I tried SGLang, vllm, and TensorRT-LLM. Under SGLang the model outputs garbage with long contexts (known bug), vllm just doesn't want to work no matter what (see errors above) and TensorRT-LLM doesn't support Gemma 3
Ollama doesn't support prefix caching and crashes when running two Gemma queries at the same time (known bug)
There's more to choose from, don't worry
https://github.com/vince-lam/awesome-local-llms
bwaa
Non open source maybe LM Studio but I forgot the backend
whats that for?
im trying to get sdl to work
Still rocking on that onedrive I see
yep
whats sdl?
imma try aan olde rversion of sdl to see if that works
input output shit for web
ahhh
"Simple DirectMedia Layer, a cross-platform development library"
Did you download the SDL mingw or the windows one?
Can you link the tutorial?
onedrive
today i will show u guys how to set up sdl2 with sdl image
sdl3 may be too new tho
ill try sdl2
👆 windows just is annoying sometimes with libraries and installing stuff
yep
especialy whent the main source is like.
so use this command "sudo xyz linux blabering"
and im like, hey stop that linux witchcrafty i dont have that
Thats a sign
windows may be annoying sometimes but its also easier in others
i just want to paly games with decent compatebility
if i have to type choco install to do anything i get angy
Like what? Productivity suite? That's just software incompatibilities
drivers for my wifi card, UI is just easier, other drivers for stuff I have is windows exclusive, some games I have are not able to run on linux at all even with proton, etc
ui is what
Driver is distro specific. In arch case, sudo pacman -S linux-firmware can already get you very far. UI design is to taste. And you can customize your UI in linux however you like. The rest is again, software incompatibilities not ease of use
aight i got it to work with some weird trickery
shoudl work now
never used makefiles before
You make one yourself?
your life is about to change in the positive or negative direction
i mean its not oo dificult with a tutorial
the website for my drivers for the wifi card and other components I have say they are windows 11 specific, and ive been using windows since I was 3 years old so im fairly used to its UI, id prefer to keep it how it is, also a lot of libraries I use for vision AI barely or require a ton of work to get working, or might just not work on linux so that is ease of use
i find it annoying how i cant use the paly button now
unless there is a way to link that to the makefile
Alright, seems still pretty basic for now. Because later down the road if things gets complicated, I suggest learning to use meson
Yeah, that's what I mean with it is not ease of use issue but software incompatibilities. Windows specific thing is a pain point in linux but not linux fault to a degree. But why in gpds name do I have to jump through hoops just to get a development environment working in windows is another matter. Also, if you use OpenCV for vision then doing it on linux is indeed easier. And you can customize KDE Plasma to looks and feels like windows
hblang hashmap update
To an extent of course
pushing and popping 5 million elements in 670ms
why does mambaforge get the tag miniforge 
Whats your target again?
TF opencv is not easier on linux im sorry, when you first pip install on linux it lacks any and all CUDA support and GPU acceleration, and using GStreamer is a nightmare, also fricking TensorRT it a pain to set up. and setting up RealSense SDK is even harder than all of those
you can make those do just about anything, you're looking for "vscode launch configurations"
also this https://github.com/microsoft/vscode-makefile-tools?tab=readme-ov-file#debugging-and-running-targets (assuming you have the extension this repo is of)
ideally as fast as rust but it's very much impossible
considering rust uses simd, which hblang doesnt have
there's an easy solution here 
what is simd even?
so, you know vectors?
Single Instruction Multiple Data
i though it was something with arrays
mhm
oops caps
mhm, rust is doing this in 290ms
also we are now libc free in hblang
yay
14kb binary

Thats good enough then isn't it? Rust is only 2x faster with SIMD so good enough for not using SIMD
oh my sweet summer child 
also btw the advancement to 600ms from 1100ms came from uh
some really strange bug that i found
in the perf data
where libc memcpy was using 44% of the execution time

if he can figure out python package management with cross compilation surely he can figure out make right 
it wasnt even the hashmap
but in the hasher
?????
Data -> Hasher -> Hash
why are we moving memory around in a hasher
ask this snippet
it shouldnt be a problem if the compiler optimised away the idk
but it doesnt

Lesson in optimization: Never trust the compiler lol
yeah so, i just rewrote the as_bytes function to accept owned values and left myself this note
the main issue with hblang optimisations are that they dont exist

side note we got comptime value substitution and inlining now
I'm waiting until there is AbleOS is complete. TempleOS successor
also, why is my terminal split now? 
pt output back in the tbs please
Click the trash icon near the junction or the terminal and output
doesnt seem to be doing much
Or you can try drag and drop
fun fact
the only parts of hblang binaries that were linked to libc were _start and memcpy which are both inserted by the compiler and left to be dealt with by the linker
these have now been implemented in nasm 
aight i got sdl2 working 
only took 5 hours
Now do sdl3 
no
Also, my displays are flickering like crazy after restarting
Thanks for this incredible software nvidia
"You're welcome" - Jensen
What is sdl2?
Funnily enough the hotfix drivers were more stable than the "stable" drivers
Man I love entirely disconnecting from the internet to switch from 4g to 5g
apparently the twitch outage yesterday was unrelated and they caused it themselves, some crazy timing that is
Wifi or mobile?
Considering I'm in a trailer out camping I think I'll leave it up to interpretation
So Evil did chew on the cable huh
Oh, sry I just joined the chat, didn't knew ;-;
Usually if you lose network when changing signals it's the phone
Yeah, you'd think it'd be solved by now
wifi 7 having band switching on the fly and all of that.
Yee
Its usually the phone network adapter
Usually, I doubt it in this case
Because normally the mobile towers are always on
But if he's in the middle of a camp
It could be the case
I'm just assuming something stupid is blocking signal or something for a bit
Companies don't care that much for towers in the middle of nowhere unless it's a road
Those aliens and their weird tech
solution: don't use 5g 
Especially in the middle of nowhere
i still have yet to find an actual reason to turn on 5g ngl
But how will he stream evil vods in 4k
it just drains my power
it's just a waste of battery
It seems to be behaving, it isn't like I rely on it much as I have everything I need downloaded.
twitch doesn't even support anything higher than 1080p 
nvidia upscaling (real)
CAUGHT
(same program) can you tell which one links to libc
remember when they talked about adding 4k support 2 years ago or so
Ye, 5g is kinda a joke rn
i wonder if that's ever happening
they're adding 1440p this summer
aight im going to bed. goodnight 
Judging by the time, bottom one
Apparently it's only real "use" rn is if you plan to stream in hd to a mobile VR headset or smt
you technically already can stream at 1440p tbf
its true, the second binary uses libc and the top one has is libc-free
instead twitch just gives me some fun options
wifi replacement is a legit use case
old screenshot
this means higher bitrate btw 
Wdym?
the one that doesnt link to libc is also 3kb smaller (17kb -> 14kb)
so mild improvements all around i suppose
Like, replace home wifi with open signals?
you'd think av1 and local transcoding would be very high prio considering it can potentially save them a lot of bandwidth
yeah
I'm assuming it is mobile phone system for standalone headsets
AR would probs benefit the most
13k with release mode
-Oz when 
incidentily the equivalent rust program is 365kb
when the flag to disable inlining happens, it will be -Oz enough

yes
which in turn is based on -O2 right
Oz is often smaller but can be larger sometimes
or was it O1
since its a heuristic or something

i guess it doesn't really matter either way
im still a little bit mad that the biggest performance gain recently has been finding a missing compiler peephole
600ish more points until I can post my terminal
i love how i managed to work around my pc needing multiple tries to post by using hibernation and now nvidia said frick you, we'll give you really bad monitor flickering occasionally when you wake your pc from hibernation 
how big is zig hello world
massive by default since it's statically linked
is it less than 2kb
this is also statically linked 
zig's stdlib is also a lot bigger i'd imagine tbf
Isn't rust also big because static link?
Hey bud, I think you're missing a bit
rust binaries are gigantic 
zig hello world is abt 6k
Where is the 130.8 GB of the remaining storage going huh? 😠
33k without llvm 

rip
there are probably some flags that can be set though
like using ReleaseSmall instead of ReleaseFast
that gets it down to 4.1k
yeah that helps a bit
can confirm its statically linked
also this is how a minimal memcpy looks
memcpy:
cld
mov rcx, rdx
rep movsb
ret
memmove is kinda the same but with one more jump with some setup depending on if rdi is greater or smaller than rsi
x86 is so stupid
funnily enough for no llvm one ReleaseFast + strip is about 20k smaller
how to use new linker
-fno-llvm -fno-lld
unfortunate
2.1kb is pretty unbeatable for most languages though
@opaque sigil am i doing this wrong 

pls
idk i blame glibc
All my friends hate glibc
i managed to get zig to 4.0k
4032 bytes to be exact
so slightly under
whats happening
we have nothing better to do than comparing binary sizes of hello worlds
I might be able to help
im getting surprisingly large c binaries compared to what i expect
so you want it to be smaller?
hmmm
what about custom syscalls or sm?
interestingly it makes the zig one larger
zig is supposed to do this itself
its kinda its whole deal
zig already uses musl
the point is not doing that
ohhh idk man haha
not in code, anyway
weird

amazing
sorry I'm pretty unfamilier with this so i dunno if I cant help 
idk zig is weird sometimes
konii please 
it kept completely ignoring me asking for a dynamic executable too 
real morale of the story: using hblang
who allocates in their memmove
ehe
i said they were temporary

old commit btw update
the only thing you have to do is change iteration order, good luck 
who up allocating they memmove
omg meeeeee
ur delirious
its only called in three places, ergo never my problem
that reminds me, i have a 75GB movie on my phone i should probably delete
I'm sorry, what!?

That's not a movie, that's a virtual disk
bdremux will do that
If the movie isn't at least 5 hours I question the non-existance of a compression algorithm
3 hours, close enough
That is the furthest I've seen from close enough
I figured out why my custom keyboard wasn't working 
its cos 4 of the pins are dead for some reason and aren't connected to the mcu 
Is it a CPU issue, a PCB trace issue, a soldering issue
no fucking clue


Fixed all but 3 
Not a software issue confirmed, definetely hardware
one of the remianing has a loose diode, not sure about the others

why do you need a better one?
faster compilation time, and I also need it for particle/volume simulations
i got so tired of youtube's shit, so i wrote this. if you use anything that isn't windows (sorry, they just don't have the needed utilities for this), feel free to use this.
(Don't forget to edit VIDEO_DIR though)
youtube dl already exists
it literally uses that
Most of the times its switch issue. Try shorting the pad on pcb and see if the keypress are registered
Or you solder a diode in reverse
It was shitty solder joins on the pins
Yè, that happens. Is that a kit or you design it yourself?
the point of the player script is fourfold:
- Integration with FreeTube (A FOSS Frontend for YouTube)
- Reduce Bandwidth use (Plays local copy in the case of repeat viewing.)
- Effectively a "Easier to Understand" version of yt-dlp, as there are a few settings on the first lines that are... arguably more human readable than parsing command line arguments to it.
- Since it is a shell script, it runs on (In Theory) anything.
Myself
Designed and built from scratch
which was fun
by default (but this is easily changable), it downloads at 480p30, which, for most things, is good enough imho.
Nice
Slap some kailh jade box on that bad boys and terrorize everyone around you with the click
Yt-dlp ftw
Use cobalt
CUDA makes life faster, but not easier
I should change my IDE settings to not consider tqdm results as error
that is TOO red
Is that hangul?
ye
Now all of China knows you're Korean
damn right i am
anyway it's okay now, you can stop avoiding memmove on purpose
$memmove := fn(dest: ^u8, src: ^u8, len: uint): void {
if src > dest {
end := src + len
if src >= dest + @size_of(uint) {
loop if src + @size_of(uint) > end break else {
@as(^uint, @bit_cast(dest)).* = @as(^uint, @bit_cast(src)).*
dest += @size_of(uint)
src += @size_of(uint)
}
}
loop if src >= end break else {
dest.* = src.*
src += 1
dest += 1
}
} else if dest > src {
end := src
dest += len
src += len
if dest >= src + @size_of(uint) {
loop if src < end + @size_of(uint) break else {
src -= @size_of(uint)
dest -= @size_of(uint)
@as(^uint, @bit_cast(dest)).* = @as(^uint, @bit_cast(src)).*
}
}
loop if src <= end break else {
src -= 1
dest -= 1
dest.* = src.*
}
}
}
alternatively you can leave the old impl, i'll still love and support you even if your memmove is a little unconventional 
this is unfair 
unfortunately the git issues got nuked when ablehub got reset from backup
morbing
yeah i remember something about that, oh well 
Good morning
spammed in gen and gaming but not programming smh
em dash would have ticked off a bingo box
out of curiousity, how many of you guys can just take one look at an image, and then confidently say wether it is ai slop or drawn?
im not going to confidently say i can identify 100% of ai generated content since that would be statistically impossible
however of the typical "slop" content i think it's fairly easy
its quite easy for most images
Yeah same
unfortunately
WHAT
I asked for cake and it replied
you are an artist make it yourself
also that's a big embed
see image
identify objects in image
understand full image
see if its ai slop
the pipeline grows
something tourist ai something?
look, AI image 
basicly, an amusementpark in the netherlands called the efteling has a themed area with a fairytail forest.
the centre of that forest has the fairy tree.
they made it AI
every morning i wake up to this thing staring into my soul
not fairy like the creature with wings, fairy like fairy tail.
just doesnt translate well
cute llc
nah its real
lc
oh
Good
Konii I'm lazy
Ok?
Do u think art is done in 3 seconds
Management think so
Not using raylib smh
(I'm just joking don't mind me)
Check out my high effort art i 100% did myself manually
Burn it
idk what that is
Thousand yard stare. Perfect
Also, that lilac plushie looks really good, must've cost a lot 
Soulless
I guess you could call it a game engine in c
im suppoosed to do that part
180$
Honestly not as much as I expected
third world country
im not too into the plushie scene so i thought like 100 or smthn
What i paid them is apparently half a monthly wage there
It's full month here
o7
Actual o7
i just found something interesting in the perf data of slice ops
in general
200 usd is minimum monthly wage so yeah, there's that
Well it's more about it being a one of a kind, someone needs to pay for the human labour
But I guess if the labour is dirt cheap

the top one is uh
-1?
To be fair, living expense is also scaled
Food is like less than 1 usd for a full meal
so with the top one, the perf report looks like this
Yeah, depending on where it's made the price would vary a ton
of this, most of the runtime here is spent doing this instruction over and over
Time to use other profiler
here full meal is 15-20 usd
scammed
Thats wack. I could buy grocery for a week I think










I need a better cpu ngl

