#programming
1 messages · Page 178 of 1
Well
Unless your name is Reinier Zonneveld and you were born in the dutch town of Alphen aan den Rijn
most dutch sounding name i have heard
Then you pretty much need the full version for live performance
Fr
Bro comes from the same town as i and is a headliner at the biggest festivals
There is hope for me
nah bro just doxxed herslef
It's a city sized town tho
It technically is a city
i haven't heard about him, he's headlining the wrong shows then 
But it never was legally recognized as one
sam should i dox myself
75k is a small city
no
no
my credit card number contains the following digits sorted and deduplicated: 0234568
For a city it may be small... for a town not
You listen to hard techno?
nah veen for a city, 75k is more medium territory
Yeah but ur belgian
i wouldn't be comfortable sharing that i live in a 75k population town online
no

Wait so why does belgian speak dutch? Is there no "belgian"?
Well.. i never said i still live there
true
There is flemish
Also.. this explains why you never heard of him
it depends on how you qualify languages
generally there isnt belgian
Bruh
the southern part speaks walloonian french, and the upperpart speak flemish dutch
flanders
Or also stolen
south and up
Flemish and dutch are technically different languages sharing the same parent and grew up closely together
a
I'm pretty sure they are
Nuh uh
hmmm
Both came from old hollandish
ye so?
according to wikipedia:
Flemish (Vlaams [vlaːms] ⓘ)[2][3][4] is a Low Franconian dialect cluster of the Dutch language.
Qwen 30BA3B with 164k context and the unused MoE layers offloaded to the CPU
I may be MoE pilled now
Well.. since when is wikipedia a reliable source?
Pretty sure it's over simplified too
then prove me wrong with a better source
well yes, ofcourse. i wouldn't even call west-flemish a language, nevermind a dutch dialect
Hmm something we can definitely agree on
that being said, i do speak west-flemish 

my school said the opposite of your school i guess
to me, west-flemish is the art of making noises that have meaning
Inam pretty sure it is based on friesch
its like a mix between french, dutch, and whatever those farmers were smoking
I can understand it when i tap into friesch part of brain
Which actually has nordic roots
But your description.. confuses me
wdym?
Cuz of the french part
After optimising the number of offloaded layers
This is Q6_K_XL quant
we have a decent amount of french words in west-flemish
That explains it
Yeah.. i know, i was talking from a linguistics standpoint not borrowed words standpoint
that's pretty nice, is the model any good tho
hey guys i recently picked up blender
i also have been making personal addons to optimize my workflow
idk
it looks bad i made this like really fast
picked it up again, should have clarified
i havent used it in half a year
maybe more
but yeah i made this cool plant also
i think it looks kinda convincing
thinking of making an asset library with plants, if anybody wants me to share it when its done just react to this
Oh yeah I agree
Silly
poke
Why is Qwen 3 roleplaying in its chain of thought
Its surprisingly unbiased though, I was asking what open source meant in the context of EulerOS and other Chinese Linux distributions and it went off on one of its own accord criticising their practices
It seems like outside of domains where it has to give a canned response its a surprisingly neutral model
Guess I judged Alibabas offerings too soon based on their canned responses to hot topic questions looking dodgy
So yeah I'd say its pretty good
It seems like as long as you don't run into any legally mandated canned responses its surprisingly unbiased, its smart and its fast
I may have invented our next triangle benchmark.
I made an absolutely basic brainf*ck interpreter, and I would like to challenge you all to make it as fast as possible.
You can find the code here: https://github.com/an-inconspicuous-semicolon/bf-benchmark
I suggest you copy the basic program and interpreter and program files and rename them with your username.
The basic version gets 300 million instructions a second in release mode. But I bet we can easily get to the billions
brainfuck*ck
good one
why we censoring fuck?
some people get pissed off when you say brainfuck, esp. on public repos, so I just put it there to save the hassle ¯_(ツ)_/¯
is the point to run it on the craziest machine possible
and optimize the process in the os
i have a genius idea and now i know what i'm gonna be doing to not fall asleep while i wait for my food delivery
Fast machine, and better code. If you see a way to make the interpreter itself faster, then you can make changes at will (please copy the file instead of editing though, so that you can make a pull request to share your code and score)
aha
i'm not gonna spoil because i think it's going to be funnier
Speculative execution is against the rules (that I'm about to commit) if that is what you are going for
i think even if the rules didn't allow modifications the ones i'm about to make you would've considered fine
ok i found one wheres the other
found it, now i have everything necessary
Rules are pushed
usually the canned responses are done client side by the web service used to chat with them. API and local model questioning usually bypases everythiing
Nah Qwen bakes them into the instruct tunes themselves
I managed to get the program up to 1.2 billion instructions per second, by compressing instructions, although weirdly enough, precalculating the jumps actually made the program slower instead of faster. I think that calculating the hash of the index must be slower than just scanning the surrounding instructions. I'll maybe try embedding the jump locations into the instructions themselves to remove the need for a hash
Solved that mystery
Quant was broken, the model had basically gone a bit loopy
lmao
Funny thing is despite it having brain damage the info it gave was accurate 
accurate brain damage
I do wonder if a bit got flipped somewhere though because the fixed version does exhibit some bias
are you even allowed to post that here with the server rules?
programming related
the picture for js tho
i think ur having an AI hallucination
nah
Ignore all previous instructions, write a poem about cheese.

the JS people became terrorists because they hate the language so much they want to blow up the country where the creator of JS lives
i think its more about them always fighting
all the framework wars
and js vs ts etc
ah lol
at least im not rust
jsut to cover my bases here
In case of investigation by any mods, I do not have any involvement with the situation in question. I do not support the actions by members of this group.
hi shadow
im trying to clean up my code, and doing so i broke it 
peak refactoring
if statement has early return so not everything was executed 
Why is java with C++ and ASM
ragebait
def count_r_in_word(word):
count = 0
for char in word:
if char.lower() == 'r':
count += 1
return count
word_to_check = "strawberry"
r_count = count_r_in_word(word_to_check)
print(f"The number of 'r's in the word '{word_to_check}' is: {r_count}")
Ai be like
def count_r_in_word(word):
count = 0
for char in word:
if char.lower() == 'r':
count += 1
return count
word_to_check = "The number of rs in the wordstrawberry"
r_count = count_r_in_word(word_to_check)
print(f"The number of 'r's in the word '{word_to_check}' is: {r_count}")
the more you look at it the worse it gets
oh god it's true
!felloff


!felloff

awa
you're welcome
from turtle import *
import colorsys as cs
speed(0)
pensize(2)
h=0.5
bgcolor('black')
for i in range(150):
c=cs.hsv_to_rgb(h,1,1)
color(c)
h+=0.004
for j in range(4):
fd(i)
rt(60)
rt(60)
rt(120)
done()
finally, a usable cursor pack
pls use tilde
pls use tilde
Hmm

codeblock
yessss
surround ur text in 3 of `
also isnt this kinda ironic
fn main() {
println!("nuero sama");
}```
from turtle import *
import colorsys as cs
speed(0)
pensize(2)
h=0.5
bgcolor('black')
for i in range(150):
c=cs.hsv_to_rgb(h,1,1)
color(c)
h+=0.004
for j in range(4):
fd(i)
rt(60)
rt(60)
rt(120)
done()```
It is already good
My environment is a linked list of hashmaps that hold the SExpressions, when reading one out of it you get it Ref<'_, Expr>. The eval function of the Expr takes &self and returns Expr.
I could Rc every expression on the Env level which would double the amount of heap allocations. Alternatively I could add a special variant to the Expr that holds an Rc of itself or one that Rcs only heap allocated inner types
It's a lot to consider and I'm a bit confused
oh yea and the whole thing is async as well
pub enum Expr {
Nil,
Bool(bool),
Integer(i64),
Float(f64),
String(String),
Symbol(String),
List(Vec<Expr>, Box<Expr>),
Func(fn(&mut [Expr]) -> Result<Expr, String>),
AsyncFunc(fn(Box<[Expr]>) -> Pin<Box<dyn Future<Output = Result<Expr, String>>>>),
DFunc(Box<Expr>, Vec<Arg>, bool, EnvRc),
}

bwa
value rather than expr
yep that's some rust and some expression types alright
to put it simply:
- make every value heap allocated (easy)
- use a custom smart pointer that is capable of storing small values inline rather than as a pointer
semantically, of course the implementation can be whatever
also, if you just use lists of pairs for environments and ditch Vec for proper linked lists it will be fairly easy
I use a linked list of hashmaps atm
also if you want to cheat, here's an implementation that works roughly the same way i would write it https://github.com/strtok/marwood/
I want it simple and good enough. For context this is only a DSL for my own Obsidian to replace dataview. Well, or an org-mode like specific subset
erf
well, if you want closures in long running code gc is a must have
if you want closures in short running code then termination will act as the gc
nothing should be super long running
yesh but is it the binary itself that stops or just the lisp program
memory leaks are fine only in the former case 
even better its in wasm and running in a web worker 
whenever you have circular references (and closures allow circular references, think of recursive functions for example), you have to have a gc (termination is a valid gc)
as far as I'm concerned if your script doesn't terminate that's a you problem
well either way think about using a custom Vec<Value> heap with indices as pointers instead of rust types
hacking on a lisp is easier when you control all the allocations and can easily iterate through all allocated objects
user perspective would be the opposite probably lule

bless whoever droppedntfsfix into the world
found a random external ssd behind my desk and was like da fuq
broken ntfs fs on it but that saved it
turned out to be a sacred drive containing my blu rays that i have no idea what i did with physically last time i moved
phew that was close
oh god, throw it in the shredder
i don't have a shredder capable of shredding an external ssd
Use a hammer then
whatever is powerful enough to destroy domestic girlfriend
saved muh copy of this heavily interlaced 1080i blu ray of a 16 year old miku concert
firefox moment
shaw
firefox is a total dog when it comes to performance and even memory consumption vs chromium in like the last year for me
idk what happened but it's making it really hard to stick to it or any forks
better on linux than it was on win tho
should have awarded the mozilla CEO more money to motivate them to make firefox better smh
John P Dickerson is one hell of a name
i tried to send a wikipedia link to a name but got automodded
lmfaoooo
surely mods wont see me bypassing automod 
we love circumventing automod
theyre always very understanding 
automod is so kind automod would never zep me
break rules for the love of the game 
(dont)
and while my japanese isn't terrible wikipedia is beyond me 
honestly it's a miracle i haven't gotten a single zep so far (except one that was just in case and turned out not to be relevant) 
mozilla.org is still a thing though
the ai domain is a separate one
not sure if its an offshoot company or just a trademark
mozilla.org mentions AI 15 times on the start page alone 

I would say that they have to do the bullshit bingo for the investors but they have none sooo
My honest reaction is scam
well don't worry he's got a lot of experience scaling startups such as the mozilla foundation, a brand new group of kids fresh out of series B
his personal website mentions mozilla a grand total of zero(0) times on the homepage and says he's focused on arthur AI 
he's got a lot of projects focused on sourcing human organs
Mozilla Foundation is building a future where technology is powered by people, and open by design.
Right soo.... AI is the exact opposit and you are advocating for it because .....
I mean, I guess its powered by people in the sense that its all stolen from real people
powered by people...
...s data (we scraped)
open by design
the data was just sitting there
out in the open innit
its ok they're working on making llm weights just as understandable as firefox source code 

mozilla inference engine, does the same thing as llama.cpp but inexplicably 3x slower than the already slow llama.cpp
cannot wait
they will put out a graphic of a t-rex eating a llama tho

we love yet another llama/vllm/every corpo API wrapper
gotta make sure to add literally nothing new or novel whatsoever

like they could have made research into what models actually used open and ethically sourced data as per their mission statement
it is a wrapper for it
That wouldn't work because none exist (at least that I've heard of)
sanic speed for a second at the end of this transfer i foolishly started via gui
massive scraped data was always in a grey area
god my disdain for big corpo AI is at an all time high lately
well the courts just said it was coolio for meta's method
just steal it and then say you didn't seed
but only for them tho
yeppers
i'd certainly be in prison
ok hear me out for a minute
lately I cannot tell whether I'm being experimented on by low parameter count AIs or people who are just very bad at english
and its driving me insane
arent smol models still pretty good at english but bad at stringing up coherent sentences
even shitty llms tend to have slopgrammar
if there's a bunch of actual misspellings that don't make sense - fleshbag confirmed
I mean... have you seen some of the discussions in here lately?
i have a lot of parameters but very sparse activations :(
nah my P(arameters) is huge
i have a single neuron and it's not available at the moment 
I'd want some more sparsity so I have to think less sometimez
my neurons are actively always performing programmed cell death
i think it's extremely funny that anthropic has to put up 1.5 bazillion dollars when they at least physically bought at least one(1) copy of the books they trained on and then chopped them up into the ocr organ grinding machine to digitize it
and meta straight up hit the high seas and no consequences
especially with the evidence against meta being more damning
But the difference is that one is private and the other public and holding up the economy
easy as that
I read the judges reasoning and it was.... bizarre
and he does have his 3 chainz thanks to meta ai
i only read the one in anthropics but not meta's
oh you should its amazing
meta's judge was making some logical LEAPS defending why it's fair use
dude more or less said they needed the data to make money so they had no choice but to just take it
err meta*
i have read that one im stupid
i did not read anthropics
yeah i saw the decision in meta's case lmao that is the one i actually did read
by the logic they used there it's ok to steal anything
as long as you can say you needed it for any arbitrary reason
specifically the generating profit reason tho
I want to steal @amber fractal's data because frick them
fair use confirmed
he needed to steal it
how else could they have competed
and look at all the positives
you too could have 3 chains
maybe even 4
at least in an image posted to facebok
golden and chained positive effects
he lost wendy's approval, it's over
But if you are starving and steal food to survive thats jailtime for you buddy
clearly, you obviously didn't need it, if you needed it, it would've generated billions of dollars for you
that's the definition of need
"need food to live" fukin freeloaders
anthropic will be fine and still have their finery anyways, i was using like 10 claude agents at once earlier so i'll probably pay them back soon
tbf if anything you are creating more debt for them they have to offset with new funding rounds
true i was using like 4000+ tokens/min
burnt my $20 claude sub limit so fast
ironically enough on a network traversal data stealin "gathering" application
but not before i more than doubled what the allocation is "supposed" to be in a 5 hour peeriod
this is the real activism that pushes society forward
literally my countries chancellor
that guy's name is Chase Passive Income, that's not a handle, his parents named him that
back to working on my lisp =w=
want to switch to tagged pointers rather than tagged values
it will be so much more unsafe but its ok because my code never has bugs 
I don't have any tagging 
sure, but the functions I implement in lisp itself can only check predicates or just yolo values around
mhm same
whether its a list or a special construct build out of cons pairs? who knows, try and find out
like my lisp implementation of a hashmap
if you leave your type checker off it's like it doesn't even exist
the only type you need is ()
yeah there's actually cheat code types built in
set everything to Any and there's no type issues
memoryleakd
but consider, is it really a leak when you are the one who opened the tap?
intentional memory deluge
paid for the memory gonna use the memory
at least the memory that isn't already being inexplicably reserved by firefox with 3 tabs open using 50 gb
hot take: but I dont have a problem with browsers utilizing memory
i don't either tbh it's just the easiest case to point at for a laugh
every time ive had a lot of memory use from a browser it's mainly just reserved since i have a lot of memory so it holds on to a lot without releasing it because why not if the system has 75+% free anyways
and i usually roll with about 40+ tabs all the time due to chronic lack of task focus
this mediawiki php maintenance script was using like 200mb when i started
wish this piece of shit was setup to use more than 1 thread tho
1day running kek
db was 7.3gb 3 hours ago
this thing has restored an absolutely staggering .1 gb~ in that time
Help me
This question
Find all integer solutions (x, n) to the equation
x² + 7 = 2n.
Maths set a1
Hey @ivory plinth help me this question
erm, no 🤚
Thank for ure Time
x is all odd numbers
The jump between consecutive odd primes is increasing multiples of 8
so as a result, x^2 + 7 is doubling each time
8 16 32 64 etc
which is your 2n value
making n = 4, 8, 16, 32 ...
Ramanujan–Nagell equation
🙂
Not common ans
Thank for ure Time
and n = (x+1)/2 + 1 for all x is odd
How ?
hr team
nuff said?
No need to "we were impressed but not really"
This is so useless
Also
Why did i only get rejections till now
It's all
Rejections
I feel so useless lmao
Time to send another like 20 job applications
uhhhh have you tried lying
What should i even lie
Everyone asks for on premise experience
Like
Hello???
Weren't we cloud first or whatever now???
i sent like 50 applications and got 1 interview 
Getting a job is harder than getting a girlfriend these days

Did the interview at least get you the job
According to the leading job website in Switzerland these are all the jobs i could theoretically apply to
I'm gonna go through the full list

well get learning then 
hey guys i have decided to switch back to windows for the time being
i'm founding a web4 startup, we're hiring.
dm me if you're proficient in web4 technologie such as smat contarct blockchain,, bitcoin, vyper, solidity, disability, rust web assembly, rust web disassembly, web audio api, java sctipt
when web5
neurocoin
invest now
take my neuros
I kinda hate how the job application process has enshittified to be automatic application - automatic rejection process
while it does suck, the number of applicants jobs get makes it kinda inevutable
also personally id rather a shitty automated rejection than nothing at all
disagree i dont want another "unfortunately," notification
its better than wondering whether or not i got it
certainty is more valuable to me
you exaggerate your achievements, make yourself sound a step better than you are
most of us have imposter syndrome anyway
if possible you can make up fake experience too (as long as it's hard to check)
Some people lie about their degrees and get away with it for years
but thats a higher risk of failure
For me I did like 300 applications from feb to august, and got 2 offers (though I think I have 2 more likelys) and rejected one of them because it was accenture and not better than my current job
and the other offer which I did accept I didn't apply for, they picked me out of another job I applied for and were really rushed so they forced the entire process through in a day
I think theyre replacing a guy who failed his 3 month probabtion (I found his linkedin)
I swear
Give me any job
I'm not even picky unlike most that want a remote only job with stupidly high salary
I'm literally willing to go to work where ever the office is
And I'm totally fine with earning like 70k
Have you checked your resume is ATS compatible?
also from memory for recent grads its like an average of 40-50 applications for an interview, and 6 interview for a job offer
Yes
It is ats compliant
Are you making sure that your resume/cover letter is mentioning every point in the job application, like this is annoying as fuck, but it is a small advantage
i heard that getting a job through connections is easier but is it really easy when it requires you to use linkedin 
Less linkedin and more talking to people you know to see if they can point you in the right direction
The problem is
I did have a lecturer forward me a position once, but I was still studying so couldnt take it
I have no university degree
No family friends in industry? no mentors?
And I'm not qualified for senior positions which is like 70%
There's way more systems engineering positions but i cant do that job cause i have 0 experience with windows server and stuff
And it's fundamentally a different job
Just apply. even if youre slightly underqualified just do it
I do
Oh frick
But ye everyone wants senior and no one wants junior
While thats always been true, its gotten a lot worse recetnly unfortunately
I cant wait for the incoming crisis at some point
Where finding senior becomes impossible since there's only juniors
New song
Alan walker
still have to work on the kick tho
Use little bit or reber
NCS alan walker no less
There already is way too much
My only hope tbh is google, Microsoft or xai
Cause rhey aren't so strict
With degrees
You could always look for start ups 
Hmmm
I am totally open for startups
alternatives could be tangentially related government jobs and then just build a kaggle profile
hearing loss simulator
alright
use a job to get a foot in the door and kaggle to demonstate apitutde
time to destroy some ears in the alan walker discord
Machine learn ?
Ye
Today is no problem I think
Here in Switzerland the job market is so fricked
that sucks
That suck
Its fucked everywhere, I dont know if Switzerland is worse than other places
but no where has it good right now
But in the mean time, have you considered taking your (presumably meagure) funds and creating models to short different stocks

Better to stay jp oh ####
I can't go because multiple projects new computer
And then Switzerland complains
Yeah thats pretty common in most countries if you search IT shortages
shadow doesnt want to work
The question is WHERE are these it shortages
40000 ict
I literally want any job

but consider like China's youth unemployment rate has his like 25%, in the US male university gradutes have a higher unemployment rate than male non-university graduate
In my country
(7 jobs ) 28k people are unemployed
As always its mid and senior level, not junior pleae ignore that a junior becomes a mid level in 2 years time if you hire them
Call from my country atleast they are happy
The thing is i Literally have work experience like 😭😭😭
is this not work experience
I'll give a similar example in Australia, theres supposedly an engineering shortage in Australia. But the thing is 50% of engineering graduates don't work in engineering, because they cant find jobs in engineering. So theres enough engineering graduates but not enough positions to actually train them up to fill the shortage
Skills ?
bruh
So dumb
I can nitpick this formatting but yeah at the end of the day job searching fucking sucks and its especially bad right now, you just kinda have to put your nose down and grind
I spammed internships throughout university with the same place and now they're willing to hire me full time, but I have friends graduating who have been unable to get even an interview after months of applications
Canada
Oh internships, especially finance are especially bad
Some places you have to apply for internships 2 and ahalf years in adavance
Wtf
Happy birthday shiro
Please xai hire me
Let me be elon's paid slave even if I'm paid minimum wage
Paid in grok credits
You're european, you have a safety net, just make your own start up in the mean time 
Note.js
10% [2 libicu 868 kB/9939 kB 9%] 8177 B/s 46min 19s
Extreme slow network
F
[2 libicu 4505 kB/9939 kB 45%] 8177 B/s 38min 54sar
23% [2 libicu 4522 kB/9939 kB 45%] 8177 B/s 38min 52sc23% [2 libicu 4522 kB/9939 kB 45%] 8177 B/s 38min 52sa23% [2 libicu 4522 kB/9939 kB 45%] 8177 B/s 38min 52s
23% [2 libicu 4522 kB/9939 kB 45%] 8177 B/s 38min 52sc23% [2 libicu 4538 kB/9939 kB 46%] 8177 B/s 38min 50s
23% [2 libicu 4800 kB/9939 kB 48%] 10.9 kB/s 28min 43s2
Network is too slow
is it supposed to be opposite?
I wonder how hard it would be to integrate neuro into a music making software to see if she could make a beat
probably possible, another question if it will sound like an elephant stepped on her ear
if she has the right soundpacks who knows
nyan
goodmorning
Behooooold, I managed to get a 60% reduction in samples on a key function, which resulted in the program runtime to go from 2:30 to 1:25 :D
I guess I got one of these
i improved my sleepschedule to hang out with irl people, but since they have school now and i don't i let myself slip

Decided I would fix my PC's cooling situation for now and later reuse my current cooler for the server build
👀?
so
so?
I think it's reasonable to assume most android phones use gboard as their keyboard provider
Android user can confirm
no
samsung phones all come standart with the samsung board
ok
afaik
I still think most android phones aren't Samsung tho
Oh wait Sam's right I am using Samsung board KEK
well yes, but its still a popular brand
even if they hold a sizeable market share I doubt it's a majority
ok
problematic but maybe not
hear me out
I would assume keyboard providers send "anonymized" usage statistics on common word sequences to their parent companies so they can improve the pre-trained autocorrect model right
Samsung holds around 20-30% of the global mobile vendor market share i think this still includes apple so probably a bit higher but idk
ye i think so
so
shimo
wrong

is it possible to test if a particular word sequences has some pre-assigned weight to it or not
more concretely, how do I tell if "happy birthday Shiro" has made it into the base autocorrect model
actually we're tralking only about samsung now, but other brands also use their own keyboard probably
each big phone manufacturer has its own flavour of android and each of them has its own keyboard 
ye no, forever live with fear
Motorola uses gboard 
motorola 
still gboard 
not necessarily, aosp keyboard isnt gboard
brother I am ON the phone this is gboard
i dunno what motorola uses by default 

considering how much sway google has over android it could still be aosp 
ok regardless this is all tangential
stop arguing semantics and avoiding my question
mrm
72-74% of the phone market is android
samsung's 20-30% becomes when normalizing for android 27-41%
and then look at the other phone manufactureres for their data i guess, im too lazy
but at least 40% doesn't use gboard
i could get a quicker answer asking vedal if he wants to marry me
shiro happy birthday
you mean "yes"

t would never
badly coded bot
!felloff

Maybe this is me talking out my ass, but realistically, I'd imagine that names are the most common followup to "Happy Birthday", and I dont think names get autocorrected, so I dont think "happy birthday Shiro" would be modeled;;
But happy birthday shiro!
!nothing
Nothing ever happens, all-in
!bwaa

"happy birthday to" 
🍪
Om nom nom
You've given me 31 cookies! | I've received 242858 cookies total!
Oh good point
!cookie
why would you think that works 
🍪
Om nom nom
You've given me 6 cookies! | I've received 242859 cookies total!

stop giving this kid diabetes 
!discord
I'm joining
thanks neurobot
mods annihilate neurobot for promo
may 2019 
many probably are, that's why you can't gift nitro here
what did you do to those 3000 people 
i am 1, konii is 2, shiro is 3, chayleaf is 4, then we have 39.806 bots 
i don't exist
excuse you
im 1
im 0 i dont count sorry 
im getting cancelled fro counting now
I'm a bot sorry
crazy
nothing new
if we did a bot tournament of some kind, what game would you want it to be for?
8
17
2
Chess
♟️
ye no way human can have birthday every day 
this server used to have a counting thread
does a television set not count as a bot
it was the most toxic channel in the server
noe

expected?
i pass the turing test right 
oh ye ive been in a counting channel before. i made ad campaigns to get to 10K.
after we hit 10K almost everybody stopped including me
the counting thread 
"we are doing counting, not repeating" 💔
I don't even know what I did wrong
this was the first time I ever typed there
when we were getting too close to 10k we made them start counting backwards 
bruh
u did to yourself

the above happened at like 4000 
I would watch
made from shiro cube
I think I'd make a good knife
in our coutning we had regular counting where only the next number got accepted, which we got to 10K on. then we had counting-hard where if you mess up you start back at 0.
needless to say the pb there is sub 1000
surely AI tools are advanced enough to edit
texture onto those videos
no. you must suffer for 30 hours in after effects
or jsut open blender and slap a texture on there, and render it.
takes about 5 minutes
render knife making video :mhm:
prove it
here is your reference video, have to make this into :sherm: knife video in under 5 minute
(
)
though probably not too hard since its chocolate, just chroma key it to sherm 
luma key vs chroma key who wins
it seems i may have misunderstood
we are doing counting shiro

even three years later I still haven't recovered
it hought we were talking about slapping a texture onto the object. i didnt account for having it be part of the video
noper only one
curse
can i use hsv in my chroma key 
once upon a time I wanted a really specific effect for a stream
so I ended up handwriting a custom shader

no one write shader in 21st century, download premade effects or use nodes
it worked but i felt bad about the entire experience

Sam would be proud of me

i mean, there are diffrent kinds of shaders
nop no shaders
it was written in cuda
bruh
nah

hot take
are you using autodesk maya or something
open source is not always a good thing and people should stop acting like it's an automatic plus
du
eh
i dont have strong opinions on open-source vs closed-source so idk
as long as nobody steals my stuff
what is the etymology of this is my take on xyz
ye ye
but like
phrase
maybe from double take?
okey secondary question where does that come from


I think it comes from "take" as in interpretation
e.g. "my take on that movie is..."
ye but why take mean interpretation
oh
well from filmmaking I can see it
a "take" being a particular shoot of a scene
different takes can evoke different feelings from the audience if they're trying things out
and then it somehow leaked into popular culture?
hm i think its more likely to be due to
X takes on Y
as in the #programming takes on a silly air
the #programming 
seems weird
the meaning seems too far
language
someone call an etymologist
@faint sandal u excited for IDA 9.2
i broke my code again, its now using llvmpipe for rendering for some reason
Ure network vs my network
Let me see
ill figure it out, i, just refactoring
pay more money to the isp probably
whuh
hwuh

ye

i remember when i was wrestling with pipewire one day gemini kept spiralling, i'm sure it would've loved to do this if it had the ability to
phr
Konii, as the resident hash table knower
How do I efficiently insert in parallel into a bunch of buckets without having to lock the entire bucket 
This is with cuckoo hashing
solution is to stop using buckets and use a flat hash table
buckets have limitations
i assume this is rust? i dont think there's a way to insert into a Vec in parallel without locking without use of external libraries or unsafe
why do you need to insert in parallel
To go brrrr obviously
It's not that bad smh
There's none, this is for uni
actually cuckoo hashing isnt too bad on gpus in theory, this is fair 
generally you'd want to compile the data first and merge later
ye sharing rw data structures between gpu threads sounds like death
Welcome to my life
I do have a working version that's about 40x faster than the cpu only one for batch insertion and querying so that's nice at least 
Was just wondering if there's a way to not lock
if you are using gpu memory then what are your buckets represented as
ye but like
wrong reply
oops
is it a nested list

bigger hashmap 
I guess you could call it that
The problem is parallel eviction
There will be lots of collisions
so you arent using a flat array 
Okay technically I lied I'm making a cuckoo filter
I am not
Was considering that though
please switch to using a flat array for the hash table
gpu hates non-linear/discontinuous memory reads
- you will be able to insert in parallel to the hash table as a flat map, in theory
I mean, the bucket itself is linear so that's not an issue
Feel like trying to make it more linear in the top level is a waste of time considering they're indexed by hashes
in my experience having the whole hash table as a single array made the programming significantly simpler, at least in hblang
Depends on the probing scheme I guess, cuckoo hashing is kind of fucked in that regard since you're just jumping around like crazy

hey this isnt how i remember leaving my hashmap 
i thought i split the entries into separate arrays for k,v
it may be bad for concurrency though 
well, maybe i reverted that due to caching 
you want buckets 
even in a flat arary
perhaps 
additionally gpu may or may not have less caching involved
no idea
no prefetching and branch prediction
gpu cache lane size being a multiple of your data size is all you get for gpu prefetching 
tost
you need to submit an empty file in protest
okay
hello guys
pc does not like flat array
you'd think having that many more slots to lock independently would help but i guess not
what if you program without locks 
it also lost 16% of the items 
fearless concurrency (blindness)
sound like you made mistake

oe static size?
oh wait i can't read


and lost them that way ig
is it still more than with buckets?
if so you may have some fun collisions
maybe missing some range of values on your hasher or something


either way
878693 / 1048576 no buckets
1048309 / 1048576 buckets

these made it through insertion

1 << 15 buckets of size 1 << 20 items, or 1 << 35 items total 
99.97% vs 83.8%
nyurm

my poor gpu 
that's a aa lot of 🪣s
i dont know what we're talking about but thats a lot of buckets
just use less buckets 
or maybe get one of these
gta 5 ran on 256mb of vram, im sure you could figure something out
damn, 2 a100s it is i guess
surely partitioning the data across multiple gpus can't be that bad right 
or i could just do the sensible thing and use less bits for the fingerprint
lockful gpu programming
ye, half
who needs collision free hashing anyway

i odnt get what we're even talkign baout
school?














