#programming
1 messages ยท Page 5 of 1
print keyword my beloved
good morning
As in too new or old?
Trying to get video permissions by typing a lot of words!
New Codex-1 model 
Why am I watching this I have no idea what any of this is
It be like this sometimes in #programming
Topos
Mor
Mor
Looks really interesting so far
imagine using python<3.13
i use 3.12 for webgl, there arent many wheels for 3.13 yet
you can use for reference
https://github.com/cern-nextgen/reflmempp
https://github.com/alpaka-group/llama
but at the end, its easier to convert to some common format
Morn! 
the what?
all the docs (that i have seen so far) uses QWidget instead of QOpenGLWidget
the latter does GPU accelerated graphics, and is basically plug-and-play acceleration if you used QPainter
QOpenGLWidget provides functionality for displaying OpenGL graphics integrated into a Qt application. It is very simple to use: Make your class inherit from it and use the subclass like any other QWidget, except that you have the choice between using QPainter and standard OpenGL rendering commands.

huh
interesting
so opengl on qt framework?

i thought c++ would work with opengl pretty well, didnt know you'd need a lib for it
altho qt is not the exacxt same thing as c+, but i cant remember the difrence
in general mixing display-type frameworks is a terrible idea
especially with Qt, since it has its own memory management rules and all
i mean opengl isnt exactly available on all platforms
depends on if you count webgl, but even then support is iffy
fairs, but i wished there was subtle text in the guides i've used for qpainter that at least talked about QOpenGLWidget 
free acceleration 
Doesn't Qt use ANGLE anyways? so what API you end up with in the end is still ambiguous even if you write opengl commands.
i don't know enough about Qt to say anything, all I can say is by painting on QOpenGLWidget, my CPU consumption dropped significantly
oh is ANGLE windows only?
yeah it is
it's irrelevant to me then 
Makes sense, no neet to translate opengl to d3d on anything else 
clearly there needs to be a linux-only drop in replacement called DEVIL
isnt it angle as in a corner? not the diety?
MESA already made that and called it Zink 
missed opportunity
ANGLE - Almost Native Graphics Layer Engine
black mesa 
theres also GLOVE https://archive.org/details/glove-src
can we standardise shaders already
use the opengl ones, i likse those
my own subjective opinion is based so ye
SPIRV is just that actually!
its a shader IR thats fairly easy to convert from or to anything else
fair, but you do have to type 200 lines to get a triangle
SPIRV is not meant to be written by hand
its an IR
IRs are made for computers, not humans
its like complaining that the LLVM IR printout takes up 20 A4 sheets when you print hello world
if i understaand this grapg correctly, spirv is also used by opengl?
its not
GLSL is probably the best for compatibility, now that D3D12 supports spir-v as well as dxir
oh, openCL, not GL
also this still doesnt mean spirv is used by opencl
the graph isnt about what uses what
then what do the lines mean?>
its about formats and conversion tools
ohhh
GLSL -> glslang -> SPIR -> SPIRV-Cross -> GLSL 
You joke, but we do that at my job ...
https://www.khronos.org/spirv/
wait, when you click on the SPIR-V Language Ecosystem picture it changes
Best workflow for web
thought of the most basic, useless project ever and managed to do it in 3 and a half hours (with insane carrying from chatgpt for getting out of loops and looping inside another loop)

chargtp
the small and evil and intimidating horses
they will steal your soul
Breaking out of nested loops is hard shit. 40 years of c++ and they're still deliberating on how to solve this ...
date of birth is a checkmark 
I still don't even know what a nested loop means, is that just a loop inside another loop?
ye
oki just making sure
goto exists for a reason
also not sure whether to feel accomplished or "meh" after this, since a lot of the difficult parts were done solely by chatgpt
you should not code with chatgpt, only use it to explains tuff
I kinda did a bit of both, I'd try one thing, it didn't work, I'd ask why it didn't work, then I'd ask what it would do instead
then ask how that worked

rust solved this (it can be argued whether its the best solution, but it gets the job done)
I take it that I used the robot healthily then
ye
still kind of leaving room for myself to learn some things that it gives me before asking what something means too, which ig is an added benefit
like in my case it was what the lines "return", "break", "continue", "return()" all meant and how they're different
return()?
yep that's a thing
I think named loops are probably the best way (even for c++ there is a proposal currently) if you want to avoid goto at all costs. But the syntax really is almost the same as solving it with goto so I don't know what benefits people expect 
I wish people would stop with their almost irrational fear of avoiding goto at all costs sometimes
Even the proposal for named loops admits This code is clear enough, but the goto is socially problematic.

imo the problem with goto is the unclear raii semantics (they might actually be clear in the standard idk) but goto is normal in some contexts yeah
Just don't do raii and that's not a concern anymore 
there are not many people I would trust to use goto and not fuck it up
isn't it a basic functionality of c
If you can't use goto to break out of a nested loop then idk you might be a lost cause
Jumping out of a scope will cleanup all objects in that scope properly. Jumping past object initialization is not legal iirc and should not compile.
The only time I saw goto being used was the default cuda cpp file from visual studio
Or using it for a simple defer
Combine goto with dynamic programming and malloc
I would first question why you ended in a situation where you feel the need to use it. Most likely you already fucked up at that point
That's fair I guess
They used it to handle errors, i.e cleanup data before returning
But yeah I haven't ever used goto myself, I forgot it even existed
Isn't that what normal error handling is for? By for example bubbling up the error?
I guess, but the example basically had multiple if statements that goto a deallocation before exiting the function
its not always desirable to bubble errors up
the linux kernel uses plenty of goto but I trust them to know what they are doing
If they returned without deallocating it would've caused a memory leak because I believe the memory was allocated in the very same function (I might be misremembering)
And I guess it's slightly more inconvenient to slap those deallocations in every if statement
Yep that's basically it
The poor man's go defer
Is nice sometimes
thanks for making it the correct defer
i'll never not get annoyed when thinking about go and defer
So if I return something from the arena it's basically garbage by the time the function exits? Surely the compiler prevents this? 

based
I like compiler guardrails made by smarter people than me. Idk 
usually it'd be
since you're adding them to a stack and then popping them off one after another
that's how it works in zig at least
well, not literally since the compiler just throws the code where it belongs but that's the idea
can't you just print to verify
ah

that's why i assume it was possible
Is it legal to place the defers above the variable they reference, since they get called afterwards anyways
no
why are two of those pascal case
and the rest snek case
ah
i guess that's true
fair enough
my random commit messages
@typeInfo when
just implement all the compiler built-ins while you're at it pls
this is peak code

honestly? not really no
i don't usually do one file per struct i'll group them
i guess if it's a big struct then sure
it is convenient but idk
feels weird
Gemini Ultra
true
i don't think that's really an issue, checking overlap for an array of pointers makes no sense in the first place
yeah
could just throw a compile error if you pass pointers i guess
is there a way to inline asm
wdym i can't write arbitrary bytes into the executable wherever i want 
speaking of which, jai public beta when
literally 1984
JIT my beloved
welp, i have become a contributor to the openGL wrapper im using
i wonder what the #programming ven diagram is of being an insomniac and being depressed
there has to be at least a large corelation
live
vs live

I thought AlphaEvolve was on Gemini 2.5. I'm more stunned now
Gemini 2.5 is just dynamic reasoning
Yeah and w/o that it was capable improving on the Strassen's algorithm
If u need many ideas similar to e.g. alphago or stockfish or whatever they generate a ton of samples quickly and figure out best
Aka playing the lottery
i love gambling
Humans too are just brute forcing intelligence, change my mind 
More like
Brute forcing random text generators
Till something good appears
google tpus cause nvidia is overpriced and making ur own chip is cheaper than buying like 30x markup
Moment
And now they use llms to generate better chips
lol
and not because they can just do that
because they're google
a tpu i can imagine probably costs idk 2-5k to build, vs a nvidia chip which costs 70000$/each
Hello i am a beginner to programming and i wanted to ask what are good python projects for getting better/improving at programming
When u need 100k of those, it's 500M$ vs 7B$
as i said, totally not because they just can do that because they have all the manufacturing capabilities and finances
making some small minigames is what i did
I mean was an investment
Same with the new google arm chip
oke i will be holding that in my mind
Why pay amd/intel 10k$/chip when u can make ur own
When u got idk 100M servers or how much they ever have or something and on that scale servers break 24/7
when u try to optimize and it becomes slower 
it means your compiler is smarter than you
its python
optimising python xd
I love pretraining llms which cost a gazillion dollars on the almost slowest language possible
Iirc openai uses torch
Hopefully they use the c++ version lol
I don't think there's that much stuff happening in python
Compared to the model itself
That being said, I still would rather write in cpp
i don't there's a non c++ version of pytorch
unless you want to count the original lua one
python is just a fancy frontend
there shouldnt be, its all gpu based ye
sure there's some overhead in context switching but that's negligible
or rather, should be
idk if it actually is
same thing with my opengl code, its slow but the bottleneck is my shadercode
boooriiiing
tldr: you suck git gud
Apparently python because ml researchers are math first programming second I guess
konii do you know if there's a way to not have page faults handled by the linux kernel
let's say i want to handle them myself
Someone told me once
git...
Math people are bad at coding and software engineers are bad at math, so they had to dumb it down significantly
this is one of the reasons we try not to do premature optimization at our company. simple naive solutions most of the time compile down to a very good result.

sadly the C monolith we translated in 2018 to Rust still has very C like things in it
yeah catching sigsev would unfortunately be way too late
ideally it wouldn't touch the kernel at all
i guess you could do all the memory management yourself?

yea
i wonder
if i use a thin wrapper for an array and hand out pages myself
i should be able to detect page faults before they occur
yeah but there's another issue, it'd have to run on a gpu 
handling gpu page faults on the cpu is painfully slow
What are you trying to do
surely there's a hidden register bit somewhere in x86 you can use to disable specific hardware interrupts 
are you familiar with the gh200
Can't say I am
what could possibly go wrong if you dont handle interrupts 
basically you have a cpu and an h100 connected with an interconnect to let the gpu access cpu memory directly
Oh, ai chip?
problem is
there's a single shared system page table where things get registered so if you run into a page fault on the gpu it needs to send a request over the interconnect to the cpu to handle said page fault and then populate the page table and send back the data
expeeeeensive

pretty much. and one of the latest version of Rust has a debug assert that a dereferenced pointer is not null. guess what exploded when I tried updating :D luckily it was only test code that didn't fully initialize the "main app" struct
we're at a point where fixing that pointer mess needs a comment with a counter of how many times it has been tried. so we're trying, but PO says "no customer value"...
I cannot stress enough how bad of an idea it is to mess with the interupt handller
your PC can literally catch fire
thats the worst
with real hardware on an already initialized system? pfff, what could possibly go wrong if the system doesnt react on whatever the hardware is doing
even better idea: don't feed your pc well and put more radiators instead
looks like i would need a way to hook into the mmu on the gpu which uhh, i guess doesn't work without reverse engineering but frick that
time to get a job at nvidia and yoink the internal docs 
yeah idk
i'm sure some smart soul is going to figure this out eventually
or realistically nvidia finds a way around the expensive page faults
what do you do when your working set exceeds the amount of physical memory
true, why didn't i think of that 
they should just stop caring about the economy and laws of physics and give us a 1TB HBME4 gpu tbh


what's this for
using otherwise wasted pcie lanes
when you want your m2 drive to be cozy warm
ah I figured they use all of them by now
apparently they do support gen 5 drives, that must get so toasty
they do
m.2 extension cable, free floating drive? :P
but they don't need that many, especially "lower tier" cards like this 4060 ti
wouldnt putting an ssd in your gpu limit gpu bus speed?
well yeah
but when you can already halve that without losing more than a couple % it doesn't matter too much
just stream your storage over network 24/7
not a 100 gigabit
800GbE 
too bad, ssd burns allong with the 5090
you dont even need fiber for it. You can copper that all up
We have dozens of VMs at work that are purely network based with no local image whatsoever
I have never played balatro to me it doesnt sound odd a tall. Just a single song that doesnt have too much going on
it sits in an odd stop for me. not chill enough but also not stimulating enough
konii
I caught you https://youtu.be/qmwgpPfDieI
Family Guy, Subway Surfers, Talk Tuah, KSI, TikTok, Soap Cutting, Reddit Stories and more! You'll never be understimmed with this video.
Ye this is based
adhd speedrun
attention span of an amoeba
I wonder if gen Z can sit through a 20min prog song
I once streamed myself watching 1 hour of commercials in a discord
And ads
tbh i cant
i never heard any peak song over 8 min
naah, the best songs are longer
bury the light my beloved
kid named save me
My favorite genre https://youtu.be/LmBI77ugdK4
Composers such as Arnold Schoenberg, Anton Webern, Alban Berg, Karlheinz Stockhausen, Pierre Boulez, Luigi Nono, Milton Babbitt, Elisabeth Lutyens, Henri Pousseur, Charles Wuorinen and Jean Barraquรฉ used serial techniques of one sort or another in most of their music. Other composers such as Bรฉla Bartรณk, Luciano Berio, Benjamin Britten, John ...
I would say that I love the distortion of the cello but thats just the mic peaking like crazy
it sure does have sound
nothing like because maybe
nope, we cant
I'm not the owner of the copyright, all credit goes to the author
Best music
umg owns it, so no embed for some reason
no, it isn't available even in browser
Ur county censors this Masterplan
if you dont add a random flute solo in your rock song are you even trying
thank fuck
music genre names is the only thing still beating tech bro product names
the cbat reddit guy is the living proof
some existing music genre names are:
Djent
Shoegaze
Math Rock
Cattle-Core
Black Sludge
deep psychobilly
solipsynthm
Math Rock๐ฅ ๐ฅ
๐ฃ๏ธ
And konii why are you 17 minutes in
It's working
๐ธ
โ 4๏ธโฃ ๐ฐ
๐ต
or just be sleep token and do all the genre
ๅฎถใฎ่ฃใงใใณใใฆใๆญปใใงใP๏ผmanbo-p๏ผใงใใ
ๅฐๅนดใฏๆฐๅญฆใๆใใฆใใใใใใใใใซๆใใฆใใใฎใฏๅนผ้ฆดๆใฎๅฐๅฅณใ
ๅฐๅนดใฏๆฐๅญฆใฎ็ฅ่ญใๅฉ็จใใฆๅฝผๅฅณใซๅฟ
ๆญปใฎใขใใผใซใ็นฐใ่ฟใใโฆโฆใ
ใใใผใซใซใ๏ผGUMI
ใไฝ่ฉใปไฝๆฒใ๏ผใฟใซใใทใจใฆ
ใใคใฉในใใ...
what in the fuck?
there is a Miku shoegaze album and its amazing
Parachute by ใใซP feat. Hatsune Miku
golden ratio face man 
what?
bro its 5 min
Omg a miku!!!
this one sounds pretty depressing
now im 

maybe they sing about flowers and rainbows
@SystemL is this true

eh, I got worse songs 
i have some worse songs but they would get me banned if i posted here

What a classic
i forgot im using the alpha channel too so i was wondering why it was invisible 
depthmap go brrr
clyde what are u on about
clyde go brrr
or I think discord calls it shard
But isnt that for dms only
no I had that in a server before
?
the gif doesn't embed
is this some kind of high knowledge behavior that I cannot comprehend
or is it just embed fail

fuck that got me. I was like ":O konii voice reveal"..
yes, but it doesn't embed anymore
oooh haha
I'M NOT SCHIZO
Holy same I'm glad
is 1.1TB of data for an llm normal?
and if so, how does it even train that? i dont have that much vram nor ram
i dont even think i have that much storage space
isn't training done in batches and stuff?
think of it as a sliding window through your dataset
I'm honestly surprised it's not more
idk you'd have to ask shadow or someone else
im watching shadow play balatro in vc rn so i dont want to bother him with llm stuff
btw, what about hard drive sepped being an issue?
cuz for 1.1TB ill have to spread it over multiple drives, including a hard drive
opening this in my browser works lol
ideally it should just load whatever it needs so i guess if your drive is so slow that it becomes a bottleneck that could be an issue
and it's not from tenor
would depend on how fast it can actually process data idk
IT LOADED
hahaha
so I'm gonna assume discord is routing it
which means discords server is down. which explains the previous error with Clyde
you installed a RAT, gg

its slovakian apparently
i would say close enough but idk if i should
this is the hard drive stats
bro didn't lose spin retry count 0 yet
that is indeed a hard drive 
what does that mean?
one of my oldest drives power on hours :D
almost at 50k
Whoโs Aoi
how shoudl i know?
some random anime girl program's author likes (i guess?)
she's a mascot
In celebration of Crystal Dew Worldโs 25th anniversary, we are pleased to welcome a new supporting character, Suisho Aoi! Profile Name Suisho Aoi Relationship Cousin Shizuku Suisho (17 years old) Age 18 years old (high school student) More details to comeโฆ Credit Character Design nijihashi sol...
Then, the fragmentโs position is transformed into world space, using the camera inverse
modelview matrix Mc-1 (which need not be a full inverse โ the top 3x3 portion could be
transposed only if scaling is not used).
optimizations 
altho idk if it will actually be faster cuz python
Classic python
Arc B770 coming? Intel says to stay tunedย Intel is cooking for Computex 2025 and there are now several rumored products to be announced next week. So a quick recap: Intel has confirmed there will be an Arc PRO announcement at Computex 2025. As we previously revealed, this announcement specifically concerns the Arc PRO A60 [โฆ]
ye
inverse uses a C function via numpy
transposing is also a C function, but doing only the top left uses list slicing
HOLY
i guess depending on how much vram they give and the performance i might end up saying goodbye to my 5070 ti
though ig realistically it'll be significantly slower so unless they add a ton of vram
B770 64GB variant when?
I'd be yoinking that so fast if it was real


"If you could use loops, this would be easy:
y = np.empty_like(x)
for i in range(100):
y[i,:] = np.linalg.solve(A[i,:,:], x[i,:])
But you canโt use loops."
yes you can
nobody is gonna stop you
wtf is this website?
the white image is not hdr unless i hold my cursor over it
I use loops in numpy
Vibe coded smh
i didn't notice cause i just used my trusty rss reader
"I donโt love NumPy broadcasting" ok? good for you?
a fellow broadcasting hater 
its fine for most things. for vector and matrix multiplication jsut use @ and you wont have problems
to me * should be element-wise, and @ is matmul
vector and matrix multiplication has never been the issue
when you have a handful of dimensions and suddenly you're broadcasting across 2 of them it gets really confusing 
but i also suck so
Yeah I can vouch for that
what?
like you go over 2D arrays?
i'm complaining more about pytorch than numpy tbf
ah, i don use pytorch so
they both share the same broadcasting semantics for the most part
it's pretty easy to end up with high dimensional tensors in pytorch

think of a simple rgb image for example
you'd have 4 dimensions, one for each channel + one for the batch
oh i mixed the sentence up in my head
you meant: you'd have 4 dimensions, [one for each channel + one for the batch]
i read it as: [you'd have 4 dimensions, one for each channel] + one for the batch


then use stack i guess
To be fair RGBA exists
ye, which is why i asked
as ive been spending a lot of time in shadercode, rgb is a vec4 in my head
this function is using 25% of execution time 
the np.linalg.inv(view @ proj) is especialy heavy there
i recently had to refactor some rust to not use arenas so i can use it with some other code that relies on static lifetimes due to concurrency, literally 1984
why are you creating 3 arrays every time
true
should be able to do something like
cornerX, cornerY, cornerZ = viewcorners.T
less garbage is always good
i forgot you can do that
Its indeed transposable
i'm technically gen Z so i feel qualified to answer, the closest thing to prog i like is certain radiohead songs, that's about it, most songs feel boring even if they aren't 20 minute long
it doesn't even matter whether you transpose it, just don't waste memory if you can avoid it 
technically you can http://numba.pydata.org/
i wonder if the python interpreter is smart enough to reuse the arrays allocations (unless their size changes idk)
๐คข
Needs to compile to wasm things, so idk about numba
numba is amazing
That isn't the issue

when in doubt you can also just try seeing if it runs under pypy
I haven't listened to any prog songs, but I have listened to 1:30:00+ songs in the past (and probably rn)
i know they've been working on getting support for more libraries over the years
what why would you need chatgpt for that
just answer from memory
"I'm ChatGPT, a language model developed by OpenAI. Let me know if I can help!"
Does Vedal ever come in #programming ?
sometimes
last time he sent a messege here was 16 days ago
sometimes with 4 months of a dry spell ๐ญ
he has sent less messeges here than me...
and I only started chatting here a few days ago
@olive sable do you have an example i could try getting to run via pypy btw
i'm actually curious
since pygame and pygbag both seem to build just fine
My GitHub is not up to date, but it does have a week or 2 old version
Memory leak or not enough tokens
let's seeee
The very first line of main.py has the compile arg
The rest of it is game assets and such
Ye, it's the only repo I've made so far
I can gitpush the current build in 15min
Will have take look when I get home
I know Sam is new to version control, considering some of the older versions having a bit much for github.
idk what kind of performance i'm supposed to be expecting 
it's ~60-70 in librewolf and 140-150 in chrome
with a 5070 ti
though ig good news is that pypy does seem to run
Chrome is vsync
Just running locally by running the file normally is max gpu
That could be it ye
I do get 70fps on both Intel igpu on laptop, and this snapdragon 870
it keeps crashing though i'm also trying to run it under wsl
time to try windows ig
funnily enough pygame doesn't seem to care too much when i tell it to target 3.12 but running it via 3.10 (pypy)
ye i am doing 3.13 myself
the one for python -m pygbag --PYBUILD 3.12 --ume_block 0 --template noctx.tmpl . is mainly for getting the right python interpretor in browser
You should publish some more code
this is all the code i have
i already have my hands full with school and this one
im doing bachelor independant game production in a few months
i love it when msvc comes in and ruins all the fun
oh ur an undergrad?
im still in hs
oh same
imagine
this works besides needing to slice it to 3 long instead, cuz its 4 logn rn
so ur engine is a pygame improvement?
I didn't comb through all the code I just looked at the website
not really, pygame is just the window
oh I see
its just opengl
writing opengl kernels ๐
shaders
shaders
ah
shaders are opengl kernels thooo
ye i do write shaders
but idk I don't do any rendering shit thank god
i don't think opengl and kernels have anything to do with each other
they're completely separate things
i mean sure, compute shaders can act as kernels
cuda kernels give you far more control
but i guess yeah they both run stuff on the gpu

Honestly my fav type of programming might be 6502
It's honestly a pretty good language
it's pretty fun too
programs these days don't have fun solutions to anything
..\numpy_core\meson.build:175:4: ERROR: Problem encountered: "complex.h" header does not include complex type complex float
thanks for being too lazy to update your parser microsoft
(for context)
99 ๐
it was usable
cornerX = [*viewcorners[:, 0]]
cornerY = [*viewcorners[:, 1]]
cornerZ = [*viewcorners[:, 2]]
minX = min(cornerX)
maxX = max(cornerX)
minY = min(cornerY)
maxY = max(cornerY)
minZ = min(cornerZ)
maxZ = max(cornerZ)
is now
minX, minY, minZ = np.amin(viewcorners, axis= 0)[0:3]
maxX, maxY, maxZ = np.amax(viewcorners, axis= 0)[0:3]
huge
It's very microsoft to not update in 26 years 
Python max and min functions 
what i find very weird is how i get as lot more performance when zoomed out, eventho im rendering more.
i will relook at my LOD code
you must understand, this is clearly an unsolved problem
in the meantime im gitpushing this
i like how the first time i test the compiled version it always takes 3 min to load 
i forgot to add --git to the command arg for compiling
without that one cv2 dont work
its been a week or 2 since i tested compiling, i got Could not create EGL context (context attributes are not supported) now


it failed now because of avx512
why do you detect half of avx512 as supported on zen3
can you like
not
think i fixed it, cant use core profile, need es
cant use 3.3 either, needs to be 3.0
but renderdoc only works on 3.3+...
high quality cde 
array indexes for fragment outputs must be constant integral expressions motherfucker
so many rules
Random question: I'm looking to buy a new 27 inch monitor that's at least 1440p, problem is that I can't decide between an IPS or an OLED. Since I game and do programming I would think likely an IPS would be better despite the lower quality since I've heard that OLEDs are terrible for productivity and struggle with burn in but I was curious what other people were using.
recent models are pretty decent in terms of burn in
i would say: productivity -> ips, entertainment -> oled
Coding can be a form of entertainment

I can fix this, i think
Debugging code is not entertaining lol
Or finding that the library you've been using isn't supported by anyone anymore and you have to completely change your code
it is if you learn from it
depending on what you mean by "entertainment" that is
Well, ideally you will but in the moment...
yep, fixed
vec4 tempcol = vec4(0.0);
tempcol[instanceID] = 1.0 - gl_FragCoord.z;
colour = tempcol;
damn im good
to learn from it you have to analyze your issues on a bit more fundamental level than just fixing the bug
like analyzing memory safety issues eventually lead to rust
gn
gn
q4_k_m goated
perfect balance of speed and like responsiveness
and quality
every time
i end up doing this
Qwen3 why must you process each line individually
mmm
go on requesty
click on prompt library out of curiosity
oh shit y'all bear grylls AI v2 is out

ive got 272 tests in this LMAO
Hi!
so quick question,
If I wanted to get into game development how much math and computer science would I need to understand to actually make a good game
a lot of geometry
cs isnt really needed
its good to know good software engineering practices and sometimes DSA i guess
if you look at many games' code it's really bad
the important thing is a good concept and persistence
of course the simpler the concept is to execute the better
also imo it's best invest a decent bit into basic art and music production (good games can have simple music and art, simple doesn't mean bad, making something simple that works takes skill)
I have a little bit of artistic prowess, and I have a good friend of mine thats obsessed with making music beats. So I think Id be pretty good in that regard as long as I give him a cut of money for a bit of my soundtrack. Just need to do a bit of studying in code
thanks for all the tips bhy
Doing some python frickery with a funny side project. This would be normal code if it wasn't static.
nreuo
nrewo
good morb 
i find it weird how
learning to draw improves my drawing with mouse in ms paint skills
does playing osu improve your drawing
I wish
ask @sage crag
she can sketch pretty stuff but idk if thats just from drawing otherwise or her gigabrain figuring out the wacom tablet
at least hand eye coordination must increase
i'm an osu! beginner (close to top 200k) and i still cant draw straight lines, but it should at least build the muscles and tendon strength you need for drawing
i was 5 digits at some point, i cannot draw 
just use a ruler 4head
my handwriting is atrocious as well
same
penmanship is an ancient concept, we should type everything instead
I'm a big fan of writing on paper
writing on paper > writing on pc when u need to concentrate
too bad writing on pc is faster
otherwise when u need high throughput writing pc is better
i have a epaper tablet i use for writing when i need to
it's useful when exploring math
buy typewriter
best of both worlds
the uuh resomething?
but outside of that it's raaaaaaaaaaaaaaaaaaaaaa
I always felt weird about this whole X is faster to type. Especially when talking about mechanical keyboards. How often is your brain faster than your fingers really
remarkable
yeah that
i use a supernote actually!
ceramic nibs, never dies
i never need to replace the pen

my stepmom has the remarkable pro i find it kinda a waste tho since she already owns 3 12.7 inch ipad pros
3 
disposable income
because i'm looking for something to write conspects in because it doesnt feel good to sit at my pc while ereading a book just so i can conspect it
I'm on the other end and make my own notebooks instead 
she earns multiples more than my dad lol
good for reading on beach
it looks like paper, so it doesn't destroy my attention span
also battery lasts 300 years
not on a writing tablet 
but about a week or so 
$300, looks reasonable enough, how is the response time?
Depends
If you're doing a 2D game in Godot you probably need less than if you're doing a 3D game or even 2D game raw in some language, or using a more complex engine
I feel like I'll still get distracted by that
so I'd rather use paper lol
it responds to pen strokes instantly, but apart from that it's not really responsive 
thats all i really need
remarkable propaganda
might buy it if i have spare cash
the world we could live in if the patent holder of eink wasnt a dick
just use ddg or ublock
whos the patent holder
i dont mind product ads on google but the sponsored stuff is a bit annoying
since sometimes it has what i need
E Ink corp 
wtf
how is it so expensive 
small demand + the difficulty of international bank transactions
apple and Samsung sponsoring the same thing
are they finally frds now
(i think it's scam btw
what I really need is ocr for greggs but somehow no one ever bothered to do that
you do it 
vedal can help lol
I have like 5 dozen half done projects, I'm good
I'm sure the two dozen ppl who can write gregg would be ecstatic
i found a remarkable for $450 though, i guess its the next best thing
but
pinenotes cost $400 https://pine64.com/product/pinenote-community-edition-coming-soon/
why do you remove embeds?
A bunch of the kobo ereader have note taking features and a pen
i have a kobo but its display has a bad response time
my biggest complaint though is armv7 rather than arm64
sadly I'm stuck in the kindle ecosystem which I hate but eh
i had a jailbroken pw2 back in the day
back to working on the balatro bot today
heard they are much higher to jailbreak these days
I can't be bothered tbh. its not like I would really get anything out of it anyway
koreader
ok but what for
for not having to convert everything to mobi/azw
also it has plugins and cloud sync and stuff
I only read books on it ๐คท
i mean, i also only read books on my ereader? i dont get your point
I have nothing to convert. I buy novells and they are there
yeah in that case theres no point i guess
due to amazons predetory practices often times its the only source I could even get some novels from
i get that, i wanted to read a manga in japanese back in 2016 and i had to go through a lot of hoops to buy it from amazon.co.jp and then got banned right after buying it
dont get me started. Its the year of our lord 2025 and the concept that someone might be able to speak more than one language, got forbid one of them isnt english, is still a foreign concept
Using the web based kindle reader is the only thing that works
This looks fire
yes but like every pine64 product it's basically "you're on your own, go fight it out"
this is what i gotta work with
Looks easy 
161 of clubs
I'm assuming you're getting this data from a mod or so? 
POC of cross communication of strings(arrays) between objects using geometry nodes


this actually have HUGE potentials of doing random shits, and I don't see anyone on the internet doing this. What I'm trying to do is create a raytracer(in geometry nodes) in raytracer(Cycles) and visualized in 3D space(viewport)
actually i feel like a portable typewriter with usb/bt kb support is a great concept
i'll keep at it then
straight lines
especially if it can also mirror everything to pc at the same time
bro doesn't have an optimal cursor path
super expensive tho
actually wait
i can just use a bluetooth keyboard on my ereader maybe?
I mean it's a beautiful bezier curve tho 
ok it doesnt have bt but i can build a wifi keyboard
really strange thing happens after learning / diving into functional langs
my fucking head works different

I think recreating the curve I drew with only two control points is possible
idk why I think of that

i have a wifi board for flipper zero and i can repurpose it as a keyboard host

see also one of my pinneds in this channel for more cool languages
for now i'm trying to work around with haskell for my personal projects and learning js/ts for diploma
but
ty
love that html is not on the list
jsx 
html doesn't deserve to be on the list (Nor does js)
working overtime on cross pollinating #programming and #gaming
companies love stealing data but hate data being stolen from them
Based
Depends on what you view as ethical
Imo training on anonymised scraped data is way less evil than what, say, Facebook have do with people's info
setting up the action space for balatro is a mega pain

theres so many actions
and i need to make so many masks
I think what's ethical is using data that was knowingly given by the people you're taking it from
When it comes to south africa .... 

(no AI can even manage these actions, i hope to get around to simplifying it at some point)
in my experience sending context is much harder than actions though
action space isnt too bad
in my case
i'm working without LLMs
so
thats the only way to go if you want good results in non-narrative-driven games anyway
sick again smh of course its a weekend
the cards are represented as the cartesian product of 8x53 spaces
no idea how to represent otherwise
i will figure it out
dont forget your umbrellla
Surely you have handled the case where multiple identical cards exist 
and jokers are 5x151 permutation invariant onehot encoded
yes
it's handled to be permutation invariant cause it doesnt matter in what order the cards are
and one hot encoded so it can be any
Invisible joker / Showman my beloved
is that a thing
Bold of you to assume only 5 slots exist
These are for cloning, negative is another thing
dont be like suriel trying to troll me into random shit pls๐ญ
i see 5
i will use 5
if theres more
then skill issue
isnt that for items
Game spoilers: ||Also a hidden perk does this||
I forget the max hand size that is possible. Balatro is definitely a game meant to be broken.
Jokers (And also items) can give an additional slot, but it is better to think of them as taking up zero slots
i will not make it ivisble then
so ez
idc about all easter eggs and shit possible as long as the ai can play balatro somewhat
it will figure it out

i sometimes forget not everything is about programming
just googled "timestamped tombstones"
what i meant is this https://en.wikipedia.org/wiki/Tombstone_(data_store)
A tombstone is a deleted record in a replica of a distributed data store. The tombstone is necessary, as distributed data stores use eventual consistency, where only a subset of nodes where the data is stored must respond before an operation is considered to be successful.
yea I got the apache cassandra docs too
decks can be safely ignored but negative is harder to ignore
so i need to consider this stuff right? https://balatrogame.fandom.com/wiki/Card_Modifiers
i guess i need to add n extra dimensions yippee
technically you can build arbitrarily large decks and joker stacks
but for the purpose of beating the game something like 50-100% buffer should statistically cover nearly every playthrough
antimatter is a voucher
theres 1 voucher per ante you can buy
(unless you reroll the shop of course)
note that you can beat the game with many restrictions (like not using jokers at all)
so dont worry about it too much
i would still wanna at least implement the functionality even if i dont use it later just as an exercise
since i have almost no idea what im doing
fair
its probably worth investing at least like 5-10 hours into just playing the game to understand it better
also probably download a 100% save with all unlocks so the gameplay stays consistent
true
or wait is there a button that unlocks everything but disables achievements
i remember something like that
ah good
why is there among us in the game files
What? 
There's an among us deck
hello clippy, HOW is this a complex type please explain in great detail
i'm starting to suspect humans need food to survive
Idk what you mean. I eat garbage and I'm doing just fine
if they made AI play starcraft surely this is doable
ive seen the end to this story before, same messaging ui and everything...
they got demis hassabis' and david silver's ass working on that
should be light work for an ex googler 
one of those was chess master by age 13 and the other has the ACM Prize in Computing
but can the ai pull off an idra rage quit
it probably can if you give an LLM high-level control of it










