#programming
1 messages ยท Page 30 of 1
skipping 9/10 computations 
java is a good language, ignore the haters
AbstractFactoryBeanSingletonFactorySerializer agrees with you
yeah the best java programmers i know learned it from minecraf
*Insert image of person skipping stairs here
if there is DI, it is good 
why the hate?
sleep well
like i think most of the skilled java devs learned it from minecraf for some reasons
java is well designed for what it is but it also lagged behind a lot on many features so devs had to invent all kinds of abstractions by themselves which led to billions of lines of vaculous boilerplate code
when you show me a java app i do not know where the actual logic is
oh thats why people use python
๐ญ
all i see is 10 line files with proxies and abstract interafces
well its true that python barely requires any boilerplate
structure, not logic
yeah that make sense now๐ญ
i think learning a java minecraft command course would maybe do๐ญ
java ugly code structure huh
We love oop tho
Clean backend structure from spring boot
Used on python fastapi too
"clean" 
(Its all DI)
please someone just make algebraic effects a thing
Modding ๐๏ธ
why did you send this three times
Have been that way for a while now
anyways
it's never shown that for me until this week
Did you just upgrade to the latest version?
yes, it is also an island
island?
Bruh, how...
i update every week at minimum
Or maybe you miss the optional dependency
It will not show without the optional deps
i've been using AUR's release
whats your aur wrapper of choice?
The -git version? Because the default repo already provided it
actually no that's not even the AUR release
Auto + Aurutils
it's directly from arch
Yeah, maybe you missed the optional deps. For me, it needs rocm-smi-lib. Otherwise, it doesn't show gpu usage
btop --version
btop version: 1.4.3
Compiled with: g++ (15.1.1)
Configured with: make STATIC= GPU_SUPPORT=true RSMI_STATIC=
bro i already forgot linux commands and last day i did an exam about it like i had to menorize the comands
we use kali linux more than arch like idk why
and my class loves kali linux for some reasons
the official releases do not have GPU support flag turned on
Notice! None of the binaries have GPU support, compile yourself or wait for distribution packages for GPU monitoring support!
so it's whoever recently updated the arch package
blame the gpu manufacture?
what are you talking about
Yeah, and it has been that way for a long while
you use nvidea or uhh amd?
oh ok
yeah i know
i just wondered
sorry continue
i wont bother you anymore
bye
you're fine i'm just confused by the question
i usually stumble on driver issues
thats why ,and well its about time i change my device
Nah, we just weirded out because I swear that btop on arch has been showing gpu stats for a while, but azaka only notice just now, or our btop somehow differs
That second part is really weird especially
i've not changed anything
improvement for older intel devices in the upstream deps perhaps?
guys does anyone of you poll the gpu for usage or is there a better way and im doing something wrong?
Neither do I, I just install as usual and then notice the optional deps and install that too. When I open it for the first time, its there. I do use btop before but did not notice the opt deps, and it doesn't show gpu usage
it had never been there for me
Oh, that could be it
For what purpose do you need the gpu stats?
If you just want to show it on a rice, then polling every once in a while is fine
java```
public static void public static void main (String[] args) {
System.out.print("i");
System.out.print("hate and love");
System.out.print("java");
}
for my first java code it isnt that bad
the public have to be declared 2 times
wow
i wonder how class would go
Who was it that use java here before? Isn't it not needed anymore to declar public static void class and a public static void member named main?

Wth is happening on #general-network
java isnโt c#, itโs still necessary
also in c# that just wraps your Program.cs into a main method, itโs syntactic sugar basically
Usual gen activities it seems
I know it is syntactic sugar. But the same can be said for all syntactic sugar. The point is still that we don't need to explicitly write it that way anymore
java 22 let's you skip the class declaration similar to c# but you still need to declare the main function ig
Ah yes, the main fn is still needed
I'm sorry what https://wccftech.com/samsung-executives-explains-why-galaxy-s25-edge-lacks-a-silicon-carbon-battery/
ai agent battery
????
????
Who's prompting AI for product design, have we lost all thought in our skulls?
โwe asked claude to optimize the software for battery lifeโ
I already distrust fully AI code without human, even more so for hardware stuff
nono, they put an llm on the phone which then tells you to stop using it to preserve battery
2 birds with one stone XD
and then running the llm locally negates all potential savings from that 
dont worry for that theres the NPU that we are you charging you extra for ๐
I don't mind more device having NPU. It just means making cheaper and cheaper swarm is easier from secondhand phone huehuehue
ai agents, the pinnacle of energy efficiency
i rolled the server back and it's been doing something with the drives for the past couple hours

Everyone wants more battery but sure let's make a super thin phone nobody asked for
uh oh apparently i cant pay for porkbun with btc anymore
hi ashhh
ye if u sit on ur phone u break it
how to sell more phones 101
Hi shadowww
that's probably more burgers than mcdonalds is selling these days
this would need an extremely insane amount of compute
Ngl, i have been frequenting chatgpt more the last few weeks
Mostly because if i search it on google, and google doesnt have a decent answer, i dont bother scrolling if chatgpt will just know it
Let's say average convo length is ~8192 tokens and gpt outputs 768 tokens and the model is 70B, that'd be 1.25 petaflops. At 45% utilization fp8 on h100, it'd take 16983 h100
And this doesn't account for kv cache
While if it is a 200b model, it'd take 48509 h100s
just for conpute
Selling tokens maybe
pro
the basic version is generally considered dumber
Pro cant be the only income they're using to pay for allat
idk any other methods since i dont use chatgpt now 
assuming an extremely optimistic contract with h100 for 3$/hr (with SLA, high performance networking, autoscaling costs, etc), that'd be 145527$/hr
or a monthly spending of 108'272'088$
they would need 5.41 million people with a 20$ subscription
and this still doesnt account for all programmatic api calls
- all of the researchers/engineers that demand +250k salaries very likely
- the model training runs which also cost tens of millions
they don't
they have some revenue streams but not nearly enough to cover all their costs, it's just the classic investor money afaik
eventually they'll have a serious problem
ok i'm tired of letting it do its thing i need to do something nuclear
just found out there's a current leakage in my charger bc i put the cable to my tooth and it felt tingly
๐คค๐คค๐คค eating cables
โ
through exorbitant api pricing of course
why did they rebrand their models to openai
or were the reasoning ones always openai
mm
It doesnt
They were 7B loss last year
They expect they won't be profitableuntil125B
Though that foes account for more investment costd
9uojmi9omrjuwe
i think i cant solve the systemd issue without soldering
hhhhhhhhhhh

you dont understand
i CANT be trusted with hardware
i short circuited my SBCs twice
although i guess i havent short circuited the new one yet and its kinda a rite of passage for my hardware

well, what can i say
poettering will be directly responsible for the $150 i lose when i ruin my server 
time to do it since i dont see any other options
what's even the issue? why is soldering something necessary for systemd to work???
wdym this is classic systemd behavior
you want to debug your system but systemd is a monolith so you have to solder to understand whats going on
I totally trust that systemd would have some stupid requirement, I just don't know which one it could possibly be
it fails to boot but i'm on linux 6.14 while i need 6.15 (or maybe 6.16) for hdmi and i cant use serial without jumping through billions of hoops (before i did it by having a separate power supply for my drives i think but the power jack broke so now i have to power them from the sbc and when i do that it takes up the entire gpio so i'd either have to find a way to connect all but ~2 wires which does technically sound doable, or solder)
normally i just unlock the drives manually without going through systemd to see whether mounting them fails or what happened uring the previous boot
but when i do it systemd is like "oh so these drives are mounted now cool i should shut down initrd networking and proceed with the boot process"
and if i disable that systemd is like "oh so boot process can't continue, guess i should shut down initrd networking and stop the boot process"
very cool
sounds awesome, great init system
yeah I guess you don't have much of a choice except serial then
mhm, the only other choice is installing a linux that supports hdmi on the sd card (or emmc) and debugging from there, and i mean if it had bcachefs drivers i could chroot so it could work, but it would take up so much time
if the drives are accessible individually / not part of an array then maybe you can find another power supply that can power at least the boot drive in order to update to Linux 6.15
should be easier than soldering at least
my other experience with systemd has been "ipv6 randomly breaks in systemd-networkd and restarting it is the only way to fix it"
just, not great
they are in an array, but they dont use that much power
the entire server powers off a 35w phone brick

and the problem isnt finding a power supply but that the power jack on the sata hat broke

can't use an ATX power supply?
it does have a molex connector but idk anything about hardware so i cant say if its input or output
either way, dont have a psu
if it's booting from an SD card, is the kernel stored on there?
if so, why not update the kernel using another computer?
its not that simple, i have to build the initrd and stuff
and the rootfs has modules for a specific kernel
i cant just put a new kernel on the boot partition and expect it to work magically
oh right, loadable kernel modules 
i'll try using some wires to connect just vcc and gnd but i dont have much hope
I'm out of ideas to fix it with software
have fun soldering, may your $150 SBC survive 
today I finally learn what an outer product is
(not the cross product, which is different)
I heard someone is teaching ai here ๐
oo when was the vibe coding path added
So after looking at both images, I'm just trying to make a simple chatbot
I'm a 1st year college student specialising in Ai/Ml
So I have the basics down
ollama
Ik what Neutral network is
Types of ai models (seq2seq,Rag,etc)
Well yea but that's just an interface
got serial but cant get the parameters right 
unlucky
Explain a lil bit pls
Oh also i made a 100M parameters model but it was spewing gibberish, loss was around 0.4, val loss was at 0.043
sounds suspicious
Huh?
was the loss function BCE?
its also possible the uart-usb adapter itself is not good enough, it happens all the time, and i have 3 or 4 of them but only one of them has female wires
i'll try to find soemthing
nn.crossentropy()
I may have miss-spelled it
typical causal lm loss is 3.5 down to 0.7
with 100M model having to be at ~3.3 eval loss
roughly
on standard pretraining
Well for more context i was using 100k conversational data
So that may have caused issues
Like 100k lines
Of question, answer
In CSV format
You are an ECE engineer?
idk what ece is
Some one brought it up so I just did a simple edit
Electronic and communication engineer
yeah totally (i'm doing this whole endeavor so i can avoid soldering do i look like a person who knows their electronics)
yea either u did something wrong in the model def or you extremely overfit somehow cause such an impossibly low crossentropy on eval while much higher ce on training
the only reason i do hardware stuff is so i can use linux wherever i want, i'm not touching anything else
I scrapped that whole thing and now I'm starting from scratch for the 17th time
๐
This time I'm going with RAG based model
Oh hi that's me
Ello
With this treasure i summon @median edge
Get in here
๐
I have a one big problem
That does not look good for health
I have a tvbox with android 7.1, rk3228A and wifi chip rtl8189fs
Number 
I need a newer firmware if there is exist, with working wifi and how to install it

It's easier if you have the full model number
breadboard saves the day as always
i have serial now and i can debug this surely it wont take that long
X96 (S905x) Android TV Box - Updating the Firmware and Costom Roms: Having bought one of these X96 boxes last year I always wondered if or when there will be an update available.With the thing running on Android 6.0.1 already I was not expecting monthly updates though.Sadly the box always gives an error in regards tโฆ
I love this flowchart ๐คฃ
didnt have to solder and now it turns out that systemd was failing due to a 1:30 mount timeout... come on
If it is molex from ATX standard then it is a standard connector with standard connection
that doesnt tell me anything but either way i have serial now so the rest should be "easy"
Welp, standard exists for a reason. It means you can look up which wire should be input or output or vcc or gnd. Same like knowing usb pin numbering alone could tell you its function. Because the usb has standard
however, it doesnt tell you whether usb will give power or take power, or negotiate it, you have to know the device specifics for that
Is does from the connector type. Female or male, type a or b
i know that i can tell what that molex connector is i just dont have to because i have nothing to connect to it
Honestly, any cable will do in a pinch. Just remember that 35W on a 5V means around 7A. So make sure to use a bit extra if you have thin wire
35W on a 12V means around 3A and you can even use the small cable that usb use
sure, any cable will do... as long as you have a way to connect it
If you almost resorted to soldering anyway, why not
lol deadlock
paaaaaaaaaaaaaaaaaaaain
i wanna go outside i cant stand this, i'll make the server boot the old nixos generation and hope bcachefs is stable enough to handle partial upgrades/downgrades
You finally showed up ๐
Man, how long does it take for your drive to boot
with a deadlock in kernel code? infinity
Tbf it was like a little bit before 8 when you tried summoning me and I didn't get off work till about 2am so still a bit tired
Ah ok
nice work systemd, interrupting the mount process
Ubuntu?
no and thanks god
Which is is that?
nixos
anyway the server is gonna stay down until i take a break because i can't stand this
Try running drive repair command
System md stops mounting if some file is corrupted
I am way out of touch with kernel. How do you know it was deadlock?
i assure you i know enough about linux and the boot process
two separate tasks hanging at the same time, isnt it obvious?
Oh ok , my bad
its fine 
Uhh, not really. Form my understanding separate task could hang for different reason at the same time no?
i mean technically yeah but most of the time its gonna be a deadlock
the electric bikes in china are so silent i almost got hit 3 times already just walking to the mall today
I just use linux to save resources of my laptop so I can use it for "other purposes"
we have the same problem with electric scooters in russia
anyway gtg
i mean it kinda make me have to be more aware of my surroundings but i dont mind that
good tradeoff so i dont suffocate in fumes
I think this is required regardless of walking around vehicle or not
yeah but you would normally hear the bike engine so you kinda know if its moving close to you
with electric bikes they are so silent you wouldnt hear them unless they are like 5 meters from you
not really complaining just sharing what i experienced
On a small road? Yeah, I could see it being a problem
Oh, you mean motorbikes?
I was thinking like "arent normal bikes also silent?"
Arent electric bikes limited to 25km/h anyways?
not sure what they are in china
Depends on where you live
I mean, you can go above 25 but the motor stops pushing
I think it has something to do with the vehicle classification
An e-bike is distinct from scooter style motorcycle
If it can push above 25/hr its officialy a speed pedelec here
Which you need a lisence for
If you mean an electric motorcycle, then they could go higher than 25
Yeah, this. Same here
But then again, my country is not known for law abiding citizen on the street lol


talking about bikes, my entire street has been closed of cuz the koers needs to pass here
koers doesnt have a word in english apparently, basicly just cycling race
"no parking allowed, world competition" 
I have a funny story regarding race. So our country is usually more in a festive mood at night during Ramadhan. Especially the youth. During the night, they usually perform illegal racing and it escalate down from year-to-year. It used to be that they race their car, and when the police become more strict on the crackdown, the next year they race their motorcycle, that also got cracked down. After that, bicycle, and last year it was running competition
All of them illegal as they block the road during midnight without permission
I know. And its always crowded too. Like dude, people are crowding like CRAZY
Anything for the thrill I guess. And the bets. Yeah, there's that too
heh, that looks like a little sword
Is the reverse different from inverse?
yeah
Understandable
the geometric product (which is what's being used here) is non-commutative
that is
ab != ba
(this also means there are two inverses, the left and right inverse)
a(a-1r) = 1, (a-1l)a = 1
Man, I have forgotten a lot of the math basics that I know lol
I don't think these are basics
they certainly weren't taught in college
...they are the basics of geometric algebra
which wasn't taught in college, because I didn't do a degree in math
Hmmm, I think basic to me is different than other people definition. I consider something fundamental is the basic knowledge before delving further and gaining understanding of higher level concept
I see
I think of things like linear algebra (vectors, matrices) etc to be the basics, even though that might not be appropriate
Nah, that's valid too. Most higher level math concept require a grasp of basic linear algebra
Sometimes things are just the generalization of linear algebra in higher dimension or more dynamic system
this kinda seems like a generalization of it, yeah
A tensor if you will
I'm studying it because I found out through chatgpt about rotors, which are a generalisation of quaternions
I need rotors because I want to mess around with making a 4d game, which involves rotations in 4d space
although, I also want to leave the possibility of going beyond 4 dimensions
yeah, I've seen his devlogs
they're great
I should play those games
Has he open sourced the engine? I think he said that the engine will be open sourced
Ray marching on 4D? 
XD
I mean, the vector math seems simple enough, just move further in the same direction
use sdfs to determine step size
I know, I could picture it conceptually, but the processing on the GPU...
You will fight the traditional rendering pipeline a lot, much like codeparade lol
I think it's still doing same amount of rays, right? cause ray marching is casting rays per pixel
unlike path tracing
wait, path tracing is also that
except path tracing has bounces and samples
Yeah, path tracing is shooting ray and calculating it's trajectory. Ray marching is sampling point on said ray
there's going to a limitation to the amount of stuff visible on screen at once
So path tracing on 4D will require extra number to count for the added direction, I don't know about ray marching but my guess is its the same since you need to know the ray either way
4d ray is just 4d vector direction + 4d origin
Oh yeah for sure. Higher dimension means more data to describe an object
that and you can only see limited amounts of stuff at any given instant
meaning the player has to move around a lot to build a model of the space
I should probably keep things pretty sparse to make it easier to figure out too
Welp, I can only say good luck lol. Now I wonder, does our cute AI overlord understand the higher dimensional math
After all, their learning is not limited by 3d sensoric perception like ours
I watched a great series on getting an intuition on higher dimensional objects
let me see if I can find it again
The first video in a multi-part series on understanding and visualizing the 4th dimension, from a mathematical point-of-view. We'll understand where a 4th dimension lies, what 4D space is, and learn a powerful way to visualize 4D objects.
Oh don't worry. I still have my intuition for any weird geometry
this guy finally made me understand how to interpret a terreract
I love physics so I study it as a hobby. And as a side effect, learning a lot non-euclidean geometry and higher dimensional space
I think flatland is even more impactful to grasp the basic from lower dimension. How a 2D being view 3D being
Then there's dichronauts where the minkowski metric signature is -1-1 1 1 (if I remember it right). It means you have 2 spatial dimension and 2 time dimension
for me, the cincher was understanding the stacking of cubes inside each other = depth based projection
2 time dimensions!?
Yep. Mathematically speaking, we usually model physics using 1 time dimension. But nothing stopping us from trying more time dimension lol
got donowalled by an automatic door 
I mean... you can kinda project time into a spatial dimension, with constraint of causality... but also, how does causality work across 2 dimensions!?
Huh? Like, physical automatic door? That require you to pay?
"dono"wall
right, anyway it just ignored me
Its... complicated. Here is the author blog explaining it
https://www.gregegan.net/DICHRONAUTS/DICHRONAUTS.html
Illustrations for Dichronauts by Greg Egan
well I know how I'm spending my late evening
You have to speak friend in elvish
Be careful, its a slippery slope lol. One time you read non-euclidean geometry, next time you know you ended up in random wikipedia article about theory of quantum gravity
why is wikipedia so obtuse
is it because they try to keep everything as compact as possible
or is it because nerds edit the pages to show off their vocabulary
Nothing is perfect, and Wikipedia is no exception. This page enumerates user opinions on why Wikipedia is not so great. For formal criticisms, see Criticism of Wikipedia. Much of the presented criticism is debated in separate essays: "Wikipedia is succeeding", "Wikipedia is failing", "Why Wikipedia is so great", and "Replies to common objections...
Well, I think its fine as it is because wikipedia are meant to be the starting point. As you read more jargon, you'll get used to internalizing the concept that it even saves cognitive load eventually since a concept can be described by a single word rather than multiple
counter point: if you already know the concept, when are you going to look up the wikipedia article for it
This is the only weird part tbh. The particular timeout is there because it is exactly to prevent a process from running indefinitely on boot and halting everything else
Different component will set their timeout differently, but there is also a default alternative
So if it changes on different boot for the same process, that is weird
Hmm, for me personally I love to read an article over time just to refresh my concept and reassurance
maybe I would have to do that in my thirties 
Tbh, I do that just due to sheer amount of distraction that my mind wander into lol
fair
beep
boop
FINALLY i managed to get an actual shell that systemd cant take away from me
all it took was passing SYSTEMD_SULOGIN_FORCE=1 to init and waiting for boot to fail
why is linux
so, what... every object has a direction in the time dimensions that it's moving at the same rate?? like, d = (a . t_hat + b . u_hat), such that magnitude of d = 1?
guy named torvalds:
hmm yes programming
why is systemd
can i call myself a systemd developer (i contributed like 1 line of code)
sorry, convo started here even if it went offtopic
ok its more than 1 line probably more like 10
guy named pottering:
but also I don't know about using general for this kek
Isn't that true for any n-dimension? Time is always marching and each object indeed have their own trajectory on the space-time diagram.
also, just learned that d = speed of light, and it's a real thing
I didn't know that objects had a fixed speed that was distributed across the space and time dimensions
On the usual [-3,+1] we just didn't "notice" because there is only a single possible "direction" of time
It is fun to think about lol
yeah 
it does slot in nicely with my knowledge of the fact that photons only live on the space dimensions
Now programming a 4D game with that metric is a challenge
Actually, yeah. The dude is trying to program a game in 4D like a madlad, and the tangent lead there lol
It is always fun to think about possible game that also challenge the usual status quo
the idea spawned from me wondering what gravitational lensing might look like for 4d objects
or weird topologies, like the 3-torus
Yep, and I look forward to it
...I might have to do path tracing to see that
so, that part would not be in a game
And here is the CodeParade 4DEngine
https://github.com/HackerPoet/Engine4D
Uhhhhhhhh
java```
public class script {
public static void main(String[] args){
int age=21;
boolean status=true;
String name="ekness";
System.out.println("name: " + name +" status " + status +" age " +age);
}
}
i havent learned any tutorial and i am already figuring out about java
dude its crazy if you know a language you can figure out others
my python and js experience was really something
Yeah, that's programming alright. Learn the concept and language is just a matter of "what is it called here"
Hopefully you're not too attached to dynamic types, you won't be finding any of those on Java
im not
bro i already figure out half of the basics
I personally hate dynamic typed languages
why
They're really annoying to deal with when functions might not say what their argument types should be or what they return, if anything
Also makes IDE completions basically not work
this feeling is really awesome
i see
THIS
Python is the worst offender on that list because "typing" is just a suggestion
like youy can only compile one python file
Never used it, I use full IDEs instead
At least JS is consistent in not having type
its great
Compile? Python is interpreted
wdym, i "compile" lots of python files
yeah but in getbrains comunity edition
it sucks you gotta pay
What?
Hey, it is technically possible using tools like Nuitka
You don't compile Python, you interpret it
The IDE has nothing to do with that, the interpreter works with paths and text files, not compiled binaries
"compiling" python just means compiling the parts of the interpreter you need, the needed other files, and then running it on the fly afterwards
I still prefer statically typed languages but I've kinda accepted that dynamically typed languages can work well too
LSPs solve a lot of the issues with not knowing what types stuff is supposed to be through type inference, and most languages have type annotations that can help enforce stuff
not requiring static types with a known "shape" is really nice sometimes
besides nuitka or cython, but those arent compiling pure python code, they require extra stuff to set up.
This makes it impossible to modify functions, class definitions or a few other things at runtime like you can in normal Python, but theyโre orders of magnitude faster.
technically pypy also compiles python code to machine code
same thing
no
pypy AFAIK doesn't really impose any restrictions on what you can do in the Python code
and doesn't only speed up specially designed methods like something like Cython would
pypy JITs your python code, so basically its converted to python bytecode first and then machine code
fair, but limited compatebility with cpython extensions
I mean, I think "auto", "var", etc handle the convinience of not having to declare your "super::duper::necessary::type_name"
and being explicit tends to be better for code design, or so I've learnt
most statically typed languages have implicit typing
its very easy to avoid types entirely in basic programs
hold up let me look up implicit typing again
yeah, it's basically compiled in the same way Java code is in the JVM 
except by default it completely recompiles from the source code instead of loading bytecode from class files
(though pypy can do that too I think, CPython definitely can, but that one doesn't have a JIT compiler)
ah, it is what I thought it is
incidentally since hblang now has an x86 backend we can also do this 
The NodeJS of python
most interpreted languages eventually get a JIT
their runtime cost is too high otherwise
luaJIT is a great example
Beware the pipeline lol
yeah, having a JIT for languages like that is a kind of a no-brainer
the potential speedup they offer is just insane
I learn that gradual typing exists
immediately find blog post about language author who says it's worse than both static and dynamic typing

how
gradual typing is great
it mitigates some of the issues of dynamic typing without being locked into static typing
well, you can read the blog yourself I suppose
I never used it myself
so don't have an opinion to give
found this blog too
i would love a language where you can start with dynamic typing to prototype, and switch to static for prod
TypeScript
nevermind
Lisp
The motivation for gradual typing is that you can use a language to prototype your project (or to sketch initial architectures, etc.), then evolve that using the same language into high-quality code with static types. But this is flawed in three ways: the author of the second blog predicted your motivation
flawed
the first blog post did too
The idea behind gradual typing was that it would allow you to build a prototype or simple scripts in a short amount of time using dynamic typing, then over time turn the program into a statically typed program (where beneficial).
so what exactly is the problem?
https://www.ncameron.org/blog/a-response-to-a-decade-of-developing-a-programming-language/ second post, for those who want to read it
I recently read the blog post A decade of developing a programming language by Yorick Peterse (found via Steve Klabnik). I thought it was an interesting blog post which got me thinking, and I have opinions on programming language design from Rust (it is almost exactly a decade since I
@olive sable what compiler do u use, i am in need of one
this, apparently
I love that JS isn't even on the table lol
pygbag, but this doesnt really compile it, it just turns the python file and compiler into an apk for embeding
so it can go in itchio for example
cuz i make opengl do the heavy lifting
what I'm getting from the first blog post so far:
- gradual typing is basically just dynamic typing with benefits (
) - dynamic typing is bad and unnecessary and should be avoided wherever possible (
)
For those that don't know, JS is designed in just around 10-days
and what a brilliant design that turned out to be 
oh i forgor that the GPU is the one getting cooked not the CPU
as long as you dont change too much uniform data each frame you wont be cpu python limited
depends, how far along your path are you?
I have some knowledge, but far from an expert
im confortable reading and writing python, and i can read the docs if i need to
i wont be limited by chatgpt probably
I guess, get comfortable with linear algebra?
Yeah, lisps doesn't fare better either
i havent got taught that in school yet but i learnt a bit myself
i think i can learn alongside
nice
im more of a project based learner anyways
aside from that, basic programming stuff will work too, as well as gpu specific knowledge
like being aware of latency between cpu to gpu data transfers
Lisp is a family not a specific language
i can understand those stuff
Yeah, I know that. Its just funny that a variation of it has 5 equality operator
as well as basic C memory concepts
uh, idk.
remember to unalocate the storage usage on the gpu after you're done using it.
good luck!
gpu memory leaks?
or, as a certain friend of mine would say, just let your program crash and let the os/drivers unallocate your data 
i dont think its hbvm->x86 though, just hblang->hbvm or hblang->x86
This has the same energy as sleep sort
ye... not pretty.
even my current build has gpu memory leak
i will try to keep that in mind
For each numeric element in an array, sleep for the time equal to the element, and after sleep, add the element to the result array
also, what could be a good first project for me to try?
i dont think i can allocate three years to make a game engine like Sam XD
I feel like you need to fix those before next game jam
Just like, maybe consider that, it could be a bit useful
the quick and dirty fix would be ctx.release("all") but this also unallocated my renderbuffer so i cant really use that.
i will just need to loop through each object
it took me like 6 months to go from zero knowledge in programming to making some cool little graphics in opengl with c++
im 99% sure game-jam 2 has the same issue. the game just doesnt use that much vram so nobody noticed
less than 500mb
500mb is a lot
nahhhh
true, but its the same effect 
that's a eighth of my total vram 
my brother in christ how do you only have 2GB?
-# sorry got the maths wrong
does your laptop come from the 90's?
I have iGPU...
same here ...
4gb graphics card, built the pc back in... 2020? 2019?
thing gets cooked when i play osu
so that means you're using the system ram, ie 8+GB
Yeah, but the bandwidth is shit
true
keep a list of weak references to all objects with graphics resources that need to be explicitly deallocated, deallocate something when it hasn't been used for a few frames
or just implement __del__

That's crazy
that list already exists lol
what in the
bro why did you hardcode it sdfkalkdf
Dude made a spreadsheet
i havent looked at this code in a year, im just now noticing how stupid it is that im doing multiple material classes of the same image
that seems to be just for entities, not resources
though I guess it's similar enough that it might not matter
at that point you should've created a Scene object for example
the entities are linked to their mesh
-# okay i need to go make dinner bye
each gltfMesh() puts data on the vram
I know, that kinda has to be the case, but non-entities might also have meshes or textures 
idk the details how your game engine is designed and which features it's supposed to provide
there isnt really anything else with mesh data afaik
@olive sable if u dont mind can i play around with your game engine?
there are other materials tho
sure go ahead
I kinda want to implement the resource GC idea now, but I already have work to do 
gah my gpu just died
i launched the game in my browser
my pc couldn't handle it
update this is rediculous
the game is literately updating at a slideshow pace and yet the page says its rendering at 120 fps 
my gpu nis at 100%
the game clock is slideshow but the rendering is still doing 120fps
i have no clue why the fps counter is inacurate in browser btw, i think it has to do with the python code still running even when gpu is still busy
async stuff i guess
maybe if i run it locally it will be better?
if your gpu is at 100% i doubt it, but you can try
also is this a game engine or an actual game?
mostly a game engine, but i have my test scene and stuff there
so if i want to import my own it is possible?
your own what?
my own scene and models
got it thx
New study proves that IA thinking is just a very complex patter recognition, rather than true reasoning... 
https://youtu.be/fGcfJ9J_Faw
New research by Apple on the Collapse of Ai Reasoning Models: The Illusion of AI Thinking. Thinking models, like Sonnet 4 Thinking or Claude 4 Thinking or o3, Gemini 2.5 PRO Preview are capable of some powerful reasoning, but collapse rather soon with increasing complexity. Is more thinking a possible solution for future Ai models? @Google @anth...
(still awaiting a study that shows humans have true reasoning)
(not provable, we dont)
Intel Architecture?
how is this news?
I think most #programming people know
not so sure about the median ChatGPT user
I know, I'm just tired of seeing this kind of story, or the classic "AI is so bad at thinking, it fails at (some task that humans also frequently fail at)" people need to get their act together
?
The failure is that the flaw remains after scaling the model 10s or thousands of times, or with all the research poured into it
Oh hey when did you wake up Sam?
Ah
almost 7 i guess
set this to 0 for patato pc's
also I finally finally finally finished the c++ side of my code with fancy IK for the robot arm

looks like im building a python library from source XD

pygame sure is a library to build from source, it has audio, io, and a bunch of other interactions with drivers. im sure there will be no issues 
they compiled SDL without Wayland support? 
also, no XWayland?

no, pygame uses SDL2 which has horrible wayland support

i have SDL3, and i worked with it on something else
oh right, SDL 3 is a thing now

weird that they didn't port the driver back to SDL 2
it sure is a great idea trying to compile at 11 PM 
man i really cant touch the compiler after 11PM im going to eep and try again tomorrow
gn chat
goodnight 
bro sent screenshots

yes
this si exactly what i was complaining about
-# wait you were a mod

anyways, the physics in dichronauts is crazy, this is totally the kind of vibe I want my game to have
like look at this picture:
that light cross is the sun in this world
not in our universe, no
the light is like that because of consequences of turning one space dimension into a "time-like" dimension
can some one help me with while loop
no way it's shiro
post while loop
System.out.print(age);
age++;
if (age>30) {
System.out.println("current age: "+age);
break;
}
else{
System.out.println(age);
}
}
for (int score2=50; score2>=100; score1=score2+1){
System.out.println(score1);
}```
your indentation is off
what does age start at, if it's <100 then the loop will never fire, and if it's >100 then it'll get to the first break and stop
also, write it like
```java
code
```
oh
yeah, that while loop makes little sense
it increments age but has age > 100 as a condition
usually you'd want to either decrement age or use age < 100 as the condition
I'm crashing out right now
crash in
implementing initial values for optional arguments in my lisp somehow results in me having to change elemental Expressions and rewrite a ton of shit
H
W
are you implementing them like Python where they're evaluated as part of the definition or are they evaluated on every call
Yeah it feels very depressing, all I see is smoke
rate my json list parsing code in hblang
(json. are extraneous and will be removed later)
Will this work for me? I have a box with fake android 12.1 and real 7.1
the argument list is a list of symbols which in this context is just strings and I saved them in the S-Expr as such. But initial values are in a list itself. So I either have to start parsing it again which is nonsense or I save it all as expressions and type match
languages and rewriting name a better duo
actually in more ways than one
and don't you dare say "just make a macro for it"
sounds like the second option then, re-evaluating the default values for the optional arguments on every call
seems inconvenient to implement 
that last for loop will never run will it
but thx so much
it's definitely something
bruh
the entire day has passed and i did get access to a console in initrd but i'm in no way closer to success
probably gonna have to leave it overnight, come back to something, leave it for the next day, and then it will be fixed
having no access to my server hurts
how am i gonna talk to my friends smh
i'm slightly worried about the server stability at night since it's only supported by an empty pack of tea
well, now its also supported by an empty coffee jar and a sayodevice
that does sound rather troublesome
Why even do it like that?
because of the height difference
the way the server is connected to the sata hat requires the sata hat to be upside down to access the gpio
and the drives are quite a bit taller than the server

@olive sable shrink your Git repo 
it's ~850 MB now, mostly because of a giant apk that got replaced >40 times
also TIL that GitHub doesn't have a hard repo size limit, they just recommend staying under 5 GB and their support will go find you if your repo causes trouble 
Nah
The apk is the compiled version
I know
you aren't supposed to put build artifacts into the repo
that's what GitHub releases are for
Huh I thought there was a hard limit past which you have to use LFS
brother
there is a 100 MiB file size limit, but nothing for the whole repo
damn
Whats the harm with me keeping the apk in the thingy tho?
you have to download all the apks when you clone the repo
until it's 10 GiB and Git will slow down to a crawl
Why all of them? Cant you just get the recent one?
Thats stupid
with commit history
its possible to do a shallow clone but its better to do full clones
Isnt it too late to remove it now anyways?
since it allows git blame and stuff
that makes it impossible to actually work with the repository though 
never too late
You can wipe it from the history
git push -f fixes everything
What does that do?
force push, discarding the remote version in favor of your version no matter the conflicts
its easier to get full history though, git clone vs git clone --depth 1
either way, git's just not made for binary artifacts
in fact it can be argued it's not even made for text but it works well enough
How tho?
If you want to remove it from earlier commits too you can follow this:
https://docs.github.com/en/authentication/keeping-your-account-and-data-secure/removing-sensitive-data-from-a-repository
If you just want the repo to stop growing you can remove the file (and probably the whole build directory) in a new commit and just never re-add it
Ideally it should be in .gitignore too, that way Git won't even show it as an option to add
(ideally do the former since the repo is quite big already)

this basically rewrites the entire history and only keeps things that aren't in the build folder
you'll have to force push afterwards since you'll end up with an entirely different history
#ad
@fast pagoda
Chat im 
m
Today sucked

this model is actually insane
it's literally instant
masive wall of text
both Flash and Flash-Lite are "only" 200-300 tok/s 
it starts at like 1800 tokens
per sec
then slows down

I wonder how this compares to the really speedy providers of normal autoregressive models (Groq, SambaNova, Cerebras)
who could've guessed that parallelisation could speed things up quite a lot 
okay yeah, Cerebras and SambaNova still crush it 
but they're also more expensive and worse quality than Gemini Flash and probably Gemini Diffusion
1B model 
gemini diffusion apparently gonna cost less than flash lite i think
yeah, would be weird for it to cost more
since the speed gains come from predicting multiple tokens at once without really needing more compute per token

hyper this model is absolutely godly
hi shadowt
it does diffusion on the full file basically at once

antarctica osage == hyperblaze?????????
are we for real
you finally noticed??
you didn't know ???

gemini diffusion?
ye
i wonder how do they allowed arbitary length of generation
there are details on this page of how the model works but idk if i can share
fair
a week i guess
will gemini diffusion diffuse me a job application
surely
it's fine, no need to share yet
someone will share it anyway as access to the model expands to a wider group 
this

still waiting to gemini robotics
Access
google pls give me gemini robotics access
from the 3 frames of video that I've seen of Gemini Diffusion it seems to still be somewhat autoregressive?
like, it doesn't try to predict the last tokens of the output first, there seems to be a window of a few tokens ahead of a contiguous chunk of already generated tokens, so I'm speculating that that's also how the model can decide the output length by emitting an end-of-text token
but it might just be a performance optimization to enable KV caching, idk 
likely block diffusion
or else, it is going to be suppper schizo
and uncontrollable
I still need to read that 
but yeah, predicting random tokens throughout the entire output sequence seems... difficult
Ye
Block diffusion is basically
u generate e.g. 128 pad tokens
Then have the llm diffuse them in 32 steps
So u generate tokens 4x more efficiently
So its basically diffusing on fixed blocks of pad/diffuse tokens
the the
AR model generate a group of tokens
and then diffusion model fills out the details
half-autoregressive
helps arbitary length of tokens to be generated, instead of fixed padding
anyway....
bwaa
like
Teach llm to generate text that means the same
Not necessarily perfect 1:1 text
ask another LLM to compare the texts and output the loss 
Similar to how llm embedding model trained
begin with that
Or wait
Enc dec transformer
wait no
Idk
Would have to think about it
Meanwhile i explode
generate similar variations of text from an existing dataset, then just compare embeddings to get the loss?

wait no I'm
, you wouldn't need examples for it
but I think the idea should still work, instead of doing per-token loss you get the embedding of the output and compare that to the target
Why every single SOTA tts are all llama backbone
I find it so dumb
That or qwen
Gemma architecture W
What do you guys think, I cleaned my dad's car today
multi-token generation lol
Should it be AR or non-AR, that's the question
doxxed again
fingers in the window, text on your house in the mirror
Bro how
That text is the text on the side mirror dummy
doxxed, you live at "objects in the mirror are closer than they appear"
Follow my path of Antarctica
You jest, but wikifeet exists. Who's to say it doesn't exist a wiki for finger

that's an NA only thing i'm pretty sure
how is there text in the mirror if its on the side of the mirror?
anyways i was just joking about the doxxing, this time
last time was fr
Lol
also that's the reflection of the door not the house
Man, looking at the Neuro stream made me think. Being parasocial towards Vtuber, at least the person exists irl. But neuro?
bwa bwa bwa bwa bwa

Who tf is this
reaction spamming 
me
jova is ez
- some minecraf player
literally me
The reactions are scary atp
jova is stoopid bc i can't run meincarft
bwa bwa bwa bwa bwa
why yall bwawing like those bunies from that nickelodeon show called rabit?
bwa bwaaa bwaa
is it supposed to be cute?

what

bro what
you wouldnt get it


what tha fouck?
one of us one of us
coo coo crazy -mario 2009

OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOh
do you not see how sad the bwaa is in the emote?
the joke is bnuuy, sir
This
I did a heart with the ground turkey I'm eating haha
my set up any rating?
6-10 why is the keyboard yellow
idk it changed to yellow razer rgb software sucks
i wanted to use blue but didn't work out
its a laptop, and your headphones are hanging on your display which idk how i feel about. the screen feels a bit low too.
6.5/10
But actually I'd say 7.5/10 or sm










