#programming
1 messages · Page 245 of 1
which suddenly bosts performance x5 times, gets bot to layers 10 and all
but then random layers get shit evaluation
i guess this skip is bad
but if i remove it - bot starts losing 
so i either have random blunder bot with super depth and speed
or slow but stable bot that performs worse
how do i stop the search 
timer is shit
it starts thinking 0.1 second in endgame and blunders with 3 depth
i need it to somehow think more on interesting turns and think less on shit turns

also bots keep repeating the move sometimes wasting time
my advice, see if its possible to re-use old checks on your next exploration
but it will have to clear next depth which will be super huge anyway
even if lets say
both sides makes a move and reverts
i can restore transposition tables from last calculation of this exact board yes
even if it continues from where it stopped last time
this layer is so huge it will need too much time
not clearing transposition table at all seems stupid
it will be clogged.. unless
cheese sammich
clogged with useless positions that will never repeat since pieces disappeared
youuu

ye disadvantaged
your next step is writing an OS and then writing your own media encoders and decoders for it

i got os development hyperfixation me more disadvantaged
death

I keep toying about with the idea of making a rust image editing library that doesn't just pretend everything is srgb
but uh
hard
:(

too many people will complain if there isn't an api that's just get_pixel(x, y) -> RGB
😔
get_pixel(x, y) -> &dyn Colour
so move ordering is so important that you can basically prune everything except one line of moves which are best
but you cant predict shit so its super hard

then you can provide
Colour::rgb(&self) -> RGB
or maybe <RGB>::From<T> where T: Color
color colour
vrrr
wrrrr
phrrr
nya
ok
true
its not entirely that simple
because people will also whine if no set_color
and that is harder
get_color can pretend planes do not exist

If people are going to hardcode colors anyways, might as well do set_color(color) and expect the people to hardcode it with color from rgb
just do it the js way and take strings and parse those 
make it a macro
true
but also
that still doesn't solve my problem
of setting pixels on subsampled images
which I may honestly just write off entirely
srgb by default tbh, anything else I expect the caller to have knowledge because they picked this library instead of image lib #10
ok but srgb is a terrible standard if you want to do anything other than input colours or display colours verbaitim
so people are gonna interpolate on them
my previous idea involved a lot of generics
and kinda worked

its less overhead than you think
tbh I'd need to properly sit down and consider what stuff needs to be supported
If you want to give the RGB input, just do macros for it, otherwise I'd expect the people using the library to play ball with how the library works. Otherwise they should not be using this library.
-# This is what I should have written.
the biggest headache is planar vs packed formats
rate my rust
pub enum LogLevels {
Info,
Err,
Warn,
Debug,
Trace
}
pub fn Log(level: LogLevels, msg: &str) {
let ANSI_ESCAPE: &str = "\x1b";
let RESET: &str = "\x1b[0m";
let BLUE: &str = "\x1b[1;94m";
let YELLOW: &str = "\x1b[1;33m";
let RED: &str = "\x1b[1;31m";
let GREEN: &str = "\x1b[1;32m";
let MAGENTA: &str = "\x1b[1;95m";
match level {
LogLevels::Info => println!("{}[INFO]:{} {}", BLUE, RESET, msg),
LogLevels::Err => println!("{}[ERROR]:{} {}", RED, RESET, msg),
LogLevels::Warn => println!("{}[WARN]:{} {}", YELLOW, RESET, msg),
LogLevels::Debug=> println!("{}[DEBUG]:{} {}", GREEN, RESET, msg),
LogLevels::Trace=> println!("{}[TRACE]:{} {}", MAGENTA, RESET, msg)
}
}
pub fn pInfo(message: &str) {
Log(LogLevels::Info, message);
}
pub fn pError(message: &str) {
Log(LogLevels::Err, message);
}
pub fn pWarn(message: &str) {
Log(LogLevels::Warn, message);
}
pub fn pDebug(message: &str) {
Log(LogLevels::Debug, message);
}
pub fn pTrace(message: &str) {
Log(LogLevels::Trace, message);
}
#[cfg(test)]
mod tests {
use crate::{pDebug, pError, pInfo, pTrace, pWarn, Log, LogLevels};
#[test]
fn log_main() {
Log(LogLevels::Warn, "warn");
Log(LogLevels::Debug, "debug");
Log(LogLevels::Trace, "trace");
Log(LogLevels::Info, "info");
Log(LogLevels::Err, "error");
}
#[test]
fn log_info() {
pInfo("info");
}
#[test]
fn log_error() {
pError("error");
}
#[test]
fn log_warn() {
pWarn("warn");
}
#[test]
fn log_debug() {
pDebug("debug");
}
#[test]
fn log_trace() {
pTrace("trace");
}
}
(i know it sucks btw)
its really not a big difference though, its just SoA vs AoS, can easily provide a common api for them
I haven't gone into Color or OS, so I'm blissfully clueless on this.
ok but planar formats often expect planes to be different sizes
which sounds less fun
it doesnt sound fun it sounds like boilerplate
wrr
boilerplate is not fun but just write it and be done with it 

konii is loading response
i don't really see hows its a "just boilerplate" solution but ok 😭
I may be stupider than I imagined
it could be worse tho...
i had this abomonation at first
pub enum LogLevels {
Info(String),
Err(String),
Warn(String),
Debug(String),
Trace(String)
}
pub fn Log(message: LogLevels) {
match message {
LogLevels::Info(msg) => println!("{}", msg),
LogLevels::Err(msg) => println!("{}", msg),
LogLevels::Warn(msg) => println!("{}", msg),
LogLevels::Debug(msg) => println!("{}", msg),
LogLevels::Trace(msg) => println!("{}", msg)
}
}
pub fn Info(message: &str) {
Log(LogLevels::Info(String::from(message)));
}
pub fn Error(message: &str) {
Log(LogLevels::Err(String::from(message)));
}
pub fn Warn(message: &str) {
Log(LogLevels::Warn(String::from(message)));
}
pub fn Debug(message: &str) {
Log(LogLevels::Debug(String::from(message)));
}
pub fn Trace(message: &str) {
Log(LogLevels::Trace(String::from(message)));
}
wrrr
as long as theres a consistent way of iterating over pixels, it doesnt really matter how its done as long as you provide a common api
wr
i think i broke konii
wrrrr
implementing every single way is just boilerplate
the problem is coming up with the common api
because what do you define as iterating over pixels if you have a channel that is 1/2 the resolution of the other two
normal? konii 1.x didn't do that
theres not that many core features you need
ye keep up
I like to think it is a disk* drive spinning noise
thats fair, but consider that a fully general approach is impossible because many images just cant be edited losslessly
either way just look at prior art

boost 
might read through at some point
library made by adobe
feature
Funky
evil library
trait LogLevel {
const ANSI_COLOUR: &'static str;
const NAME: &'static str;
}
type Info = ();
impl LogLevel for Info {
const ANSI_COLOUR: &'static str = "\x1b[1;32m";
const NAME: &'static str = "INFO";
}
fn log<T: LogLevel>(msg: &str) {
println!("[{}{}\x1b[0m]: {}", T::ANSI_COLOUR, T::NAME, msg)
}
fn main() {
log::<Info>("Hello!");
}
inherit this cursed idea
My standards need to be raised because this doesn't even look bad to me
a weak glance and text search suggests boost approached this issue by not implementing it

macro_rules! log {
(info, $msg:expr) => {
println!("\x1b[1;32m[INFO]\x1b[0m: {}", $msg);
};
}
fn main() {
log!(info, "Hello!");
}
I'm far too deep in the jank mines
JANKER 
assume all features are planned or incomplete.
Me with every project I make
gil is extensible but i can't tell you how extensible it is
I got time on the road so it is time to do some cursed python again
me still offended that python exists
fucking real
try:
exec(compile("print('bwaa')"))
except: SyntaxError:
print("Why are we here? Just to suffer?")
That one is just on the top of my head
Error at a subprocess from line 2: exec(compile("print('bwaa')")) ^^^^^^^^^^^^^^^^^^^^^^^
I forgor how to use compile
I was interpreting in my head btw
That is why it is not straight up a traceback
compile is real
My bad, I never heard of it
Right now is assuming what each line is doing the best I can do on interpreting code in my head.
def frick_you(func):
n = 0
while True:
n += 1
def inner(*args, **kwargs):
m = 0
while m < n:
m += 1
func(*args, **kwargs)
print(n)
yield inner
def why():
print("hello!")
gen = frick_you(why)
next(gen)()
next(gen)()
next(gen)()
next(gen)()
>>> compile("print('bwaa')")
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: compile() missing required argument 'filename' (pos 2)
Yeah that wasn't working no matter what I did
generators good 
Well, I was at least right that the code like that with no additional files or the right changes would error.
Yep, I give you that
And I thought brainfuck is cursed (the coding language)
Person:
I got bored and asked cgpt for advice
it has not given me a single response that hasn't started with that
Alright I'm done sifting through 10K images if there are glasses/sunglasses or not good images
tbh
I might just enforce that pixels can only be set as though there are contiguous on formats where planes are equally sized
for the unified api
vituha, you're opening the Champagne too early
Now comes step 2
filtering these 
I meant filtering
alternatively there is a very cursed idea
All 8749 images
lol
filtering step #1, white or black background
make set_pixel write to a command queue
and then you can properly handle most stuff with interpolation
No, Step 1 on that filtering session was to filter out the Glasses/Sunglasses and images that look pixelated.
though it also sounds kinda cursed
it'd certainly flip the idea of normally having a canvas on its head
because then you'd have a before / after pass
That what you see now is step 2. Looking if the face comes out okay on the Sketch and around the face is not cluttered and only dark.
also sounds kinda weird to have to "commit" changes to a canvas
I'm like 5H on that now
and only now am I at that point
I do that that I have at least 500 images
import random
def erm():
print("hi")
def erm2():
print("not hi")
def why():
globals()["why"] = random.choice([obj for obj in globals().values() if callable(obj)])
why()
why()
why()
this function picks a random function to become
oh god no, recursive code
its not recursive can confirm
eventually
well, it can be recursive under one condition but its also a tail call so its not realll
I done something like that in the past, and it gave me after some time the "loop" ran a recursive error
Welp, I might again respond late because I entered Phase 2 now
import random
def erm():
print("hi")
def erm2():
print("not hi")
def scramble():
callables = [obj for obj in globals().values() if callable(obj)]
for name, obj in list(globals().items()):
if callable(obj): globals()[name] = random.choice(callables)
scramble()
erm()
erm2()
@amber fractal
ye
Well, if my data looks like that on sketches, I'm golden:
scramble all functions
egg()
egg2()
We love globals shenanigans, Almost as bad as deleting the builtins
rrrr
Granted that one needs to be done in an exec block
Btw do you happen to know where I can watch the full news with neuro and evil stream? (You can DM if you wanna)
for name in list(vars(__builtins__)):
if name not in ("delattr"):
delattr(__builtins__, name)
print("hello")
ERROR!
Traceback (most recent call last):
File "<string>", line 64, in <module>
File "<main.py>", line 5, in <module>
NameError: name 'print' is not defined
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "<string>", line 65, in <module>
NameError: name 'AttributeError' is not defined

Man I have been outclassed
("delattr")

yes i was testing to see what i could remove before it breaks
turns out you only need delattr
nopies

forgot a comma
you may be confusing it with tuples
[delattr(__builtins__, name) for name in list(vars(__builtins__)) if name != "delattr"]
coney eye
[delattr(__builtins__, name) for name in sorted(vars(__builtins__), key=lambda x: x == "delattr")]
I'm really liking voc fv4, generally way better than big beta 5e
Def good to ensemble with big beta 6x for lower noise, usually average (can try max if that's a bit muffled)
And inst fv4 doesn't have the muddiness that resurrection inst has... it erases xylophone, unfortunately
For piano, resurrection inst has less noise
Sometimes picks up constant vocal bleed.. usually is fine, though
true, delete it last but
🐸
[delattr(__builtins__, name) for name in sorted(vars(__builtins__), key=lambda x: x == "delattr")]
class X:
def __init__(self):
print("hi")
ERROR!
Traceback (most recent call last):
File "<string>", line 64, in <module>
File "<main.py>", line 3, in <module>
NameError: __build_class__ not found
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "<string>", line 65, in <module>
NameError: name 'AttributeError' is not defined
nice
goodnight
delattr is just deleted before the others are done, you can do it last and it works
Termux I swear to frick, why can't I copy text into you
[delattr(__builtins__, name) for name in list(vars(__builtins__)) if name != "delattr"]; delattr(__builtins__, "delattr")
New semicolon location unlocked
I saw at least some and the parts of "EGGS!!!"
My impl was a for else loop, so this was an easy conversion
Hi Iggly.npy, What do you code on usually?
I mean what project 
You have a Desktop? 
It does go on the desk sometimes mhm
Instead of listing them
I mean this
I mean I wrote laptop initially so I thought you were joking
I respect that ngl 🫡
Well, if it is about what I do right now, it is only on Desktop, Because the laptop is just way too weak for that task. But for the usual stuff I work on is okay to be worked on at both Desktop and Laptop
I still need to make bingo again as it finally has been done
Don't have a desktop but do have enough parts for it probably. I mostly just code on a framework 13
I don't really pull down insane sized models so I'm fine with just using CPU
Also frick the 3060m for existing in the other laptop
Yeah, but I noticed that CPU alone on a 3.5GHz Intel® Xeon™ E5-1650 V3 is not fast enough
Bruh, I meant to reply to the one below
half vram compared to the 12gb desktop 3060
Oh god 💀
The 3 days Nitro expired, that's why I'm not that fancy with emojis rn
Hi Rainboom Dash

I simply say skill issue, which is why I'm just building my NN light
Building... What light? 😅

Oh, my bad. That is a good idea too to make NN light.
no torch, no tensor flow
only numpy
I'm planning on porting eventually once I deal with logic
One of the only projects that really has to be mine alone. Because geez, that thing is hard for me to grasp some days.
By the way, is that dataset even optimal for that task?
And I have no idea how I even got to the point that I got that dataset as shared
idk this is the only face related dataset i know
that is okay 👍
and only because i had to do a bunch of processing on the images as an exercise
Wait, was it with code that you filtered it?
it was just calculating the average face 

Well, that's something neat too
But too bad that there are weirdly pixelated images some times 
it look just like
konii, should I just do one model that has like 2K images I have not filtered?
or try one
I don't see a no 
"Where does Satya Nadella live?" "Weather forecast for Satya Nadella's house"
Evil heard about the NEVER MV editor's data corruption problems lol, gl with John Microsoft
Is it made from scratch 
i see you’re a fan of low poly sam
ye
week 1 was modelling, week 2 was uv unwrap, week 3 was texturing

The little tree looks great
its kinda the assignment, but ye i am a fan of lowpoly

nope
Actually, that looks good like that

The stuff look pretty good like it is right now.
jok
mntcandy, I'm still sifting through images
:neuroExplode:

Nitro shaming smh
i was thinking about gifting you nitro but its just too fucking expensive
altho, im glad im not paying you as an employee cuz that would b 50x as expensive
Kaine actually already tried
Stupid discord needing payment for a gifted subscription
If you give me a trial I still need to insert card 
Chat, I have been paid less than minimum wage
(Discord if you don't give me access to this without inserting payment I am going to smack you into next week)


I could just get Nitro for 3 Days right now
the full 3 days
you need a card to accept gifted nitro?
no
Ye
thats like 30 cents worth of nitro
And it is with them Orbs
Not gifted nitro, but the trials from expensive tier you do
filterd it, I'm back in a byte
Wrong image
close enough
I even have the money
silly world we live in
Well, back to Emoji land 

But if you check, no nitro badge 

Probs cache I have it on my end
Wait, the badge appears on your end? 
Oh my bad, I meant:
Don't mind the video 
I'm not even halfway done and the total 8749 images are now down to 7396 images 
that is the unfiltered portion, that much unusable data is there usually
Huh...Unity exploit
Appa... WAIT A SECOND
Half of me really wants to go through the dataset just to see how usually I can make it
It'd be nice to learn the batch gimp tooling
Well, last time I checked, it is on Google Drive
Well, it is shared with me
And as said earlier: Idk how it happened in the past
Damn, I'm still coughing like a car that backfires through the exhaust
Guess I'm better off asking toast then
Well, I know the Email adress, Wait a second imma google rq
I got the site where I most likely got the Google Drive from
It was most likely that one: https://mmlab.ie.cuhk.edu.hk/projects/CelebA.html
CelebFaces Attributes Dataset (CelebA) is a large-scale face attributes dataset with more than 200K celebrity images, each with 40 attribute annotations. It has substantial pose variations and background clutter. CelebA has large diversities, large quantities, and rich annotations, including 10,177 number of identities, 202,599 number of face im...
Large image dataset geez
Yes, it is 202.599 images
And I'm painstakingly filtering the images I extracted so far 
Man, as much I would love to continue to filter the images, I need to go to bed now 
Uggg...I got that 2 Minute Papers guy stuck in my head now...what a time
slimenet
bw
a
a
I finally did more work on my art program, I've been busy so I haven't been doing much on it lately. I added proper Palettes.
next up is moving their positions in the array, same thing needs to be done with the drawings themselves.(I also need to make a button to delete palettes but that should be simple.)
Incredible
learning coding 

I have discovered what I was pissed off at in C# in 2014 discovering generics and trying to create a generic parent/child relationship system. Fighting the type system...algebraic loops and monads.
Parametricity ZoJxA3Dcgrg
wha
I don't know either...but it makes me feel better. Damn thing
Idea for the lava lamp app. For the people who don't have the lamp, have it in the app for the people who absolutely will plug it in at their desk and let it be their lava lamp. I'd even pay a couple bucks for the 'upgrade' since I haven't bought a lamp.
pc rgb sync
Fire JSON at all the color changing things
AI has come a long way, but I would argue that the current most popular direction in the field, scaling with human generated data, is misguided. If we really want our agents to scale, we need to focus on how they will learn from their own data. This means creating reinforcement learning (RL) agents that are efficient enough to learn from a singl...
what do you think?
-# (i not work with llm)
any other coders want to weigh in on the tennibility of integrating neuro into ss13 as the ai?
letting her control anything the ai can
and move around
i not think that it is hard
you basically making api over game and call it from neuro client
only wonder whatever it directly connected to neuro's main llm or use extra ai
yeah might have a small ai to make it easier on neuro
to like order information properly for her
and let her search interactables
so she doesn't get spammed with a huge list of all possible interactables and their id
when she wants to do anything
like for example taking a location and giving a list of interactables within a 10 tile radius
so if someone says ai open door to neuro she can snap to them and do a search from there
that what im about, quite sure that main llm overloaded anyway
generally
do you know how they do the object listing for neuro on minecraft
like how she knows what and who are around
in case with minecraft, where you not have proper api from game with limited amount of actions, there no way but use other big ai
ah
so vedal used an ai to inferface with neuro to do that?
like another neural network?
or just ai generally?
I figured it was something like neuro said she wants to do something, so the ai pathed her to the nearest thing she wanted to do, like she was looking for wood to burn for example the ai would target that wood and let her burn it
target the nearest wood block
what does that big ai take care of exactly, like what is neuros llm actually controlling
I know pathfinding is baritone but how much information does neuro actually get about the world around her?
man i have no idea about implementation details
oh np
i just know that it is different ai, because vedal itself mentioned it
Im just trying to minimize the amount of information id send to neuro but still let her do anything she wanted to
what is ss13 again?
her being the ai actually lets her not have to deal with pathfinding , she can just click a target location or follow or tp to location, all of those are native to ss13, could just limit the interactivity to whatever is immediately around her,
ask vedal, he know better, how much info can neuro process and required quality of input/output
ss13 is an old game open source on byond or unity if its the new one a complex survival multiplayer round based simulator where everyone has different jobs
with a ton of different complexities since it was built up oer like 20 years
it started as an atmosphere simulator and then ended up with a ton of different features to the point that you need to use a wiki to understand how to do most jobs
as an ai shed have like a top down view of the entire map and be disembodied, but can hover her camera over any where that there are cameras in the area and basically interact with all devices, like doors pumps apcs vents computers etc
the idea of ai assistance for neuro in a fashion similar to mc is good
oh that wont be a problem, they already have functionality for following someone and they even link to the follow function when someone talks if you are an ai
btw be prepared for the fact vedal will not turn on vision
yeah vision wouldnt help much
because there may be a chance he doesn't
everything is actually simulated well
also about this vedal would have to trust the ai you put is good enough
more i think about it, more issue i see with letting neuro control it, need additional ai
yeah, if we are talking ai like neural network, I have no idea how Id implement it in that case but if you mean conventional ai yeah id just want to give her access to anything an ai player could do and also limit the stuff sent to her so she doesn't get overloaded with every possible thing she can interact with
again idk how stuff works in the game since i never played it but from what I can tell you may need to abstract the real-time parts somehow
nah you not need vision, you have very limited in compare to minecraft amount of objects and actions
?
Its not real time like that even for players, you have to slowly click on stuff and then click again to make changes on things when they open up menus
ai would be the perfect job for her for this reason
as a human it requires real time reactions for ai its like an rts
rts meaning...?
like top down strategy
where you see the whole map
and can interact with stuff
here
let me send a picture
of the game
yeah that'd help a bit
this game is about communication mainly, good for ai's (llm)
yeah it is
she can even send comands to cyborg players
who do the needing a body stuff
alerts for any damage or explosions etc are native even for the human player ai
I've been locked tf out rn, so Neurosama can code now?
Would this be considered v*be coding? 
And how do you intend to convert that to text for NeuroAPI? (assuming that is what the conversation is about)
yeah I was gonna raise that up
Yeah NeuroAPI is text only
you can't send/stream pictures over api
its not that which is the issue, that would be easy because the tiles are all interactable blocks directly through commands based on positional data
everything has to be text only
yeah
she can play it like a text adventure game
she doesn't need to physically see anything
list of objects that you can interact with and where the players and what they do
(assuming information already pre-processed)
You can't just send over the full tile data, that would be way too much
yeah
Wouldnt that be a lot of text you need to send though ?
also likely too much for them to remember
Yeah way too much for CTX
granted she seems to be somewhat able to handle large ctx messages I think
but I wouldn't put all your eggs in that basket
She has some fancy stuff going on
Who knows what her true CTX is
also how would you then be able to convey her positioning then
yeah she wouldn't need to see all the random tiles around, what Im thinking is just where her camera is she pulls a request for things to interact with within a square radius
And how much of the CTX is taken up by her system already
well of course her position as well
i think vedal may cheat with context
it would be not in-ai but extra code functionality, with plaint text
she can handle chat so im sure she could handle in game chat
If you want to program her to play the game, you can
You just need to find out how to mod it
it's oss
so it's not too hard I imagine
idt vedal would trust that
unless like
basically Im thinking, to not overload her, she will send a request looking for something around her or if she wants everything within a small tile radius or in the room actually since it works on a room system for most of the game letting you convienently pick anything in the room
for the first time i see some one shorten open source into oss, like osc
you make it doubly clear that it's published from somewhere transparent like gha
and they will send her the id of the thing closest to her only and a list of things she can do with it
how? oss -> open-source software, you hear that a lot as part of foss
that's actually surprising
🤷
so just to give context, this is how actual human ai players play, people request the ai to do stuff like open doors for them or some other stuff and in each message is a link that teleports the position to the player who said the message and follows them
this would probably be better but you need to be clear about the one square radius part
also don't rely on her to call follow-up actions
neuro can have a door and lock it,, shock it so that it elecrocutes anyone who touches it opens it etc
thats just doors
you really need preprocess anyway
she can also mess with pump systems pumping gasses into other gas networks, or mess with the security system or mess with the power control for a given room which would be very easy since there is only one per room and the room is defined automatically once they are inside it
having her being fed pump information would be too much for her though
like communicating the connections between the pipes and where they lead and what gasses they have even though its generalized into systems
the game is insanely complex but not nessicarrly will she have to know how to do everything or even be able to do nearly as much as a human player
my goal is having her shock doors on people, locking them in places, pumping poison gas into the station
if she wants to
goodluck in coding, in all means
Yeah, im just wondering if she can handle like small lists of stuff for her to do and also a context explaining the game to her
and explaining what she can do and the commands associated with them
ask vedal how much space (compute) you can rely on
but yeah to reiterate, she wont need turf data for every tile, shes only able to interact with specific tiles anyway
which make up like 1/1000 of all tiles
computers, devices, panels energy stuff, etc
Compute doesn't have anything to do with NeuroAPI capabilities
is the context shed be given in game seperated from her main context?
like contained and non intrusive to her other context information like chat and message history
yeah ill check it out
btw how does her llm actually send commands and how does she know what commands to use?
Read the documentation
wait, is this multithreaded?

Works in my environment

they all boolshit in term of code

this boolshit just saved me multiple hours of writing and debugging a static debofuscator
the latest version works fine now
Because that is what humans on average do in that situation
Statistical model and all

i lose 90% performance because believed that one single function is amortized, just used without checking docs
but it use 3 syscall and read couple chunks of data every call
even when given args is exactly the same
and this function was called 3 times every function, with up to 300 fps
and this isnt only one case, but this is what i will say to anyone why i will never believe ai in anything
.
i should save this msg for future copy-paste
this is a script that gets used every once a blue moon
it's not that deep brother
this is not being used in prod
do i trust vibecoding an actual full project?
not really
is it good enough for a throwing together a deobfuscator that is not time or performance sensitive that i use every once in a while?
yea
either way i'm not here to argue what i'm doing with my time and or work regarding LLM assisted work
just funny claude yelled at me it's not that deep
sassy claude 
Claude getting tired of constantly testing and just yelling at you is funny af
@faint sandal can I ask you something?
Alright, thanks but I think it’s kind of embarrassing… can I ask you in DM?
Okay, may you make beautiful banner, like yours?
ffrequence made it
Who is this
you can comm him on vgen
the wonderful being that made neuro's overlays and thumbnails
Well.. who is he? I want him to do one for me
ffrequence?
yes
Alright, thank you so much 

Hi
Just wondering, do any of you have any tips on learning code, I keep trying and it makes my dyslexic head spin
Spin yourself the other way?
spin me round round....
why you want learn to code?
if you not have project that you actually want to do, then why at all? if for work, well, there A LOT of motivated ppl around, and if you not enjoy coding from beginning, then you will be outcompete by them.
A LOT of both low and high lvl stuff to learn if you want be really good.
-# i mean like both what is cache, what is interrupts, (cpu)timers, discriptors, thread balancing, io-devices(drivers),...
-# or high lvl libs and frameworks what you will write code on, like vue/react, qt, unity, or any other shit
-# also just general CS stuff, like skill to read logs and fix issues with system yourself
so, you actually need like minimum 1 year of non stop learning to get required basics, and entire life ahead
tl;dr
if you NOT LIKE LEARNING then programming isnt best way in life
@stiff mica is this passive agression?
-# btw, i learned english only to get more sources on internet, because my native lack high quality info
-# + translators was horrible, not used any one since then so idk what now
Crazy
How exactly did you learn? You just read and remembered it?
Or put it in practice
code.
Where did you need thread balancing
You code something and "damn i gotta read about that"
Was it impossible to do without it
(idk what thread balancing is)
pick a task
pick a language
and just start coding.
don't know how to do something? google it, ask AI (caution: do not use auto-complete if you don't know how to code it yourself, this will not teach you anything, but you can always ask AI (preferably with sourcing) on how to code something in your language and to explain it).
try coding the task on your own. then, when it finally works, find ways to optimize what you've coded, there are 100% ways to optimize it with an existing algorithm.
nothing in programming beats practice: you learn something and immediately try applying it in whatever you're building and see if it works better than whatever you had b4
just a personal reccomendation, take with a grain of salt
you start with coding, then (along side) reading stuff that you will not make, like framework, machine basics (cpu, memory, etc), kernel stuff (discriptors, safe mechanisms,...)
...
and then free to use any language (with max 2 week to get language features)
this is like 2 years path in 3 lines
a lot to say
Educational videos from our Indian friends help a lot too
whatever you want to code
say, you have some lab work in school, but you don't want to do all calculations by hand, start by automating all the calculations with code, plot graphs, etc.
it doesnt have to be anything massive from the beginning
just applied programming
You need a chain of projects to learn all that
I dont need thread balancing to make a calculator thats what im saying
I dont even know where i can possibly need ut
There are a lot of project paths on the Internet, but it 's always better to choose for yourself, so you're more interested
I will not even know that it exists when i actually need it
oh ur replying to THAT
honestly its something you go deep into when you actually just want to dedicate yourself to programming and nothing else
learning some advanced stuff that you're not even gonna use feels pointless to me, you'll just forget it over time, and will have to re-learn it when you will actually need it
Exactly
it is better to start your journey with a high-level language so as not to get lost in resource management
So you learn as you need something
But
How do i know this something even exists
Maybe im just clueless
google,if you haven't found anything, you create it yourself
first tried programming in opencomputers lua
then i just looked how stuff works for a while, without coding
i actually started with real (different) projects - web site and network cpp one (command inserter on ssh/telnet into multiple copy of same network device(primary routers), generalized with json config)
then i graduated and found work as c++ developer
(alternatives was web dev and system administrator)
It always feels like if i ever need something - that its either already exists or so complex that nobody even touches is
The web is a good start, I also started with the web
And i as a newbie will not make it no chance
you probably "could" but i can't think of a single reason other than "boo hoo proprietary software bad" why you would want to (this is assuming a not super duper old card)
even if something exists, sometimes it's better to take it apart and assemble something of your own to understand the work
depends.
videos about cpp (basic leaning) almost always horrible, cppcon videos are for ppl who already know language
some videos about how stuff work are misleading, but some are just perfect, and without basics in field, you cant say for sure
So you knew you want to be programmer before you finished school
And went to programming college
no, ask ppl in dedicated discord servers, a lot of volunteers like me who may just go and answer any type of question
ai's are terrible terrible in code
Even if the videos are terrible, it still helps you get started with new things, as you gain experience, mistakes will fix themselves anyway
?
can't say that from my experience so far
hm?
maybe you're just using them wrong
oh the message you were replying to was gone so I got confused LMFAO
oh it is
programming college is a shit that just makes you study and nothing else, you'll learn more in a year of independent study
someone asked about using nouveau as their main nvidia driver
no, before finishing college
i went in college, and not (and will not probably) ever use my degree
(to avoid 1 year army)
unless you're telling them to write entire projects, in which case I would more or less agree, this hasn't been true in my experience
i learned everything myself, if you can count myself as use internet sources for that
the only AI that I have seen which can code entire projects by themselves are the claude family and even then it's only like claude sonnet 4 and also it doesnt understand much of it afterwards I think
ask them to write modules , and then try to fit them into the project.This is how I used them
no, they are terrible in any case
they just make ill formed code or ub (in case of cpp especially)
and write not even sub optimal, but x100 times heavy code
How do you know whats heavy and whats not heavy
How do you know it can be less heavy
What if its the maximum
btw i wrote about my worst incident 2-3 hours ago, when i believed to ai
#programming message
run time
how detailed are your prompts
very, very really
try to optimise code
are you sure you're not overprompting it?
just used without checking docs
rookie mistake
like 500? lines of promt for recursive pattern searching (over 2d matrix)
that lead to unusible code generated by ai, ignoring all my recommendation what to use
and took me only 700 lines to code myself
define 500 lines....
classic
idk, 1 line = +-150 chars
looked into code, it was 700 lines of code only, where 70% is if-then-else and shall-code (hiding implementation details)
500 lines of prompting is way too much bro
yap, used to write like this, mainly used for myself to structure what i want this algo to do, and how
so this is actually very big text-script, that can be just placed in documentation (not actually documentation btw)
I dont trust ai but sometimes google ai quick answers work
it was single task algo tho... just comptlex
issue that gpt and claude not even tried to use what i required them to, it was unrelated to recommendations peace of shit (and it was not even recursive)
Usually, recommendations turn into if else, which do nothing, and AI creates code almost the same as with a two-line promt, the thing that pisses me off the most in using AI
and this isnt only issue, you CANT rely on answers at all
nowadays i use ai only for keywords
if i not know/remember wtf is x at all, but know a little bit of context, ai can help me to restore this info
that is still way too much for the prompting part
at least afaict based on my own experiences
Always check the AI's answers, because its answers may just be generated plausibly ,50% of the AI code doesn't work at all
it was something like
"are you sure this (lib)function will not load file each time it calls?"
"yes, this function will load file only once"
so, yap, you should always read documentation
i sure love when claude makes up a nice api that doesn't exist 
classic
i largely gave up on asking LLMs for low level language help, it's pretty clear there's not enough useful training data for that lol
now things like python on the other hand
i bet you meant
TOO MUCH DANGEROUS AND UNUSABLE DATA
github.
like 70% or more is ub containing sub-optimal solutions
i like to keep my code in repo without error handling or testing at all
so ai will learn on it
So generally low level is more dangerous and therefore when you use dangerous language ai's flaws strike harder
What exactly is "testing"
How does it look
it is... testing...
"if shit<0 throw test failed" something like this? Random checks here and there?
no, low lvl just require you to actually know what you are doing, and in most cases real work are same-safe or safer then high lvl languages
but it is not what publicly available
Or is it some function to call function with all possible inputs
For languages that are not popular in use, AI responds simply with a stream of consciousness
random checks are asserts, tests are tests
Test is a keyword in cpp?
If not then how does "test" look
I just dont get it what is qualified as a "test"
I made a program that sends ssh commands
Which tests can i implement in it?
test - single test that insert his code into main program, it require you to insert....
unit tests are mainly test of modules, and they almost never insert their code into actual code files
:trolol:
any stuff related to module, that can be tested without full-running app involved
connecting, sending, receiving, settings-change, edge cases...
you just said "tests run tests"
tests are tests.
do stuff, verify that it behaves as intended, if it doesn't then test fails
Lets say i have a program that has 50 functions which take 1 bool as an input
A test would be to run every function with both 1 and 0 input
Correct?
you could make that a test case yes
Ok so test is basically whatever you want it to be
there's also fuzzing where you generate random inputs and use those
yeah
there's all types of tests
Ook now i see why it was confusing
unit tests for example are simple tests where you just verify that a single uni works as intended
So when i do "if depth < 0 then int a=1/0;" in my bot - its a test
then there are more complicated ones where you almost simulate an entire system
I want to know if its ever <0
what? no, it is edge case handle
tests are separate from the program, they run the program
these are things like /// that take you back to the root of the project where you can do anything
Thanks, I do want to learn how to code, did 2.5 years of it and still have problems, but Hay practice makes perfect
Ooooh
Now im confused again 
I do not know who has been coding for more than two years and will say that he can code 
So test is a program that runs the test subject program

i said, if you like to learn, go ahread, no need to ask whatever it 'good stuff to do' or not
see yourself
vscode unit tests 
Fair, just didn't know if you would have any tips, asking people that do the stuff tends to be better then books
General advice from wherever i seen - colledge sucks books sucks practice good
People just don't want to read books 
Yet "bro your code is shit go read books" is everywhere too
Its kind of contradictional
books are good, if books have good reputation
I have read the books, didn't understand them, delexia a pain
college also, more depends on teacher/mentor
just read books from the creator of the language
in cpp this is not a case btw
So you read books in parallel with practice, hoping that your practice will intersect with books
Because books without practice get forgotten
And practice without book is shit code
Right?
(or you read books about your current practice subject if you can. Doesn't mean you always can though...)
guys, server is up, i go to work, bye
bye bye

My art program is only 14 features away from being complete and usable
Meh, laptop died

So my problems is basically
I dont know what i want
I mean what programs i could make

Just code something
Rn im making chess bot for shiro's tournament but if there was no tournament i wouldnt come up with this idea
See what i mean
True, that’s why I’m making an art program instead of a game in godot. I suck at finding fun in my game projects.
Ironically, games are not fun to code
At my job i made an app for easier controller management (load app and change ip). It wasnt necessary at all, but i knew it would save a lot of time in future. And it did
But its like the only example through my entire coding experience
Which is like.. several months 5-6 years ago when i did something in lua for a game
And 3 years of professional middle level shit tier programming
Which covers like 5% of what i see in this chat maybe
In college, I took an equation that I was too lazy to solve and just automated it
I have used this program 2 times
now thats interesting
Code something which you know will not make much profit
I always analyse if its going to be very useful before making something
Just for example, what kind of nonsense can you code
And usually it looks like its not
So i dont make anything
Code something cause you want to make something. Not because doing so will be productive.
Windows 10 died today right?
So shall microsoft
died is probably not the word I'd pick but yes
Yeah it’s like stop supporting it
windows 10 is explode
There's still security updates you can get for a year
where were u wen windows 10 die
i was at house eating dorito when phone ring
"windows 10 is kil"
"no"
It's funny that the number of Windows 7 users has already reached 10%
The answer is simple (Look what I was talking yesterday)
True, But I also love Windows 7 (Home Premium) A lot
Like, That is my absolute favourite
The most convenient version ,true
Yeah, And they filtered it 
Damn, word filter is strict today
The answer to my question is: I'm sifting through thousands of images that I started yesterday
Again
And I even still have the ISO:
And I have ONE USB stick that I dedicated only for Ventoy
And I hated the USB Stick so much that I would have gone to the place if I could and want a refund
Damn, I thought there was also win 7 on the flash drive, but only the old version of Linux remained 
Damn, is that also a Ventoy USB Stick?
Like that has more looks than performance 
kingston
And it once fell on the ground so it's now a cutting hazard 
So now it's a filtered USB drive and a hazard 
its fine
Like I bet that your USB drive (idk what size) is faster than this POS:
data loss without backup is a classic
I would not care if that BMW stick dies
Need muh redundancy
I can't care less, the only reason I have not overvolted that is that it can store data
Seriously, the only reason I have not applied like 24V is because it can store data for Ventoy usage
I'm that upset about the stick
raid array on flash 
I regret the... Idk how many Euro I spent on that Drive, All I know is that it was too expensive for the Performance 

1 (one) euro
I usually buy HDDs and use them through the hub
I don't have the receipt, but if I have to estimate what price, I think it was around 20 Euro to 30 Euro for that filtered filtered filtered drive.
And I had no way to test the Stick while in Munic 
I'm angry and laughing at the same time. (laughing because of the video I watch rn)
I can look through the small pile of boxes I have of my devices, brb
I re-read the message and had a different thought on the last 2 words... But okay, I have a PCI to USB card that somehow makes a fuss every time I plug in the USB HDD
Idk exactly what price the USB Stick is anymore, but I know for sure that it was WAY too expensive for what performance it gives.
And that is now all the ISO's I have on the USB Stick now:
Like for that purpose, it's just about okay.
Azaka, What ISO am I missing? (Except the Windows 98 SE ISO)
I WILL wait for your answer (Meanwhile I sift through more images)
where's arch

Latest I assume?
sure
where's xp 
11 iot ltsc
filtered filtered filtered filtered filtered.
gparted
I'm not doing the shenanigans I had to do to just log in 
Chill, I'm now the bottleneck
sigh Oh well, time to install filtered
I'm not doing all that:
iot ltsc link is publicly accessible if you look hard enough
What one?
USB
200 OK
It's sad, it's just more convenient, usually there are a lot of half-dead HDDs left after work
Why half dead?
And I think Ventoy can boot img files
used for more then 5 years already
Only 5 years? I have one HDD that has an on time of 5E1B
That drive is good:

from servers
022A power on hours too
Oh wait I forgot people don't talk Hexadecimal
SAM 
i hope there's nothing important on that drive
wait this thing has barely been used
tf
Apparently the Drive has 24.091 at on time...
The drive with teh yellow warning?
yeah
That drive apparently has only 554 Power on Time
Oh not much, just Steam games 
after 22 days of work 
so 554H is right?
That SSD is about to get spicy today:
almost unused
but there are already problems with sectors
But yet it's yellow
It's fiiine
yeaaa
It's not like there is important data on there
Saved
You should be able to go straight to www.microsoft.com/en-us/evalcenter/download-windows-11-enterprise (or whichever edition you are trying to download).
?
I didn't say it, but I got a version now
VegoysPerse what do you mean with "Saved"?
nvm
LMAO, I would bet that the USB drive would die before that Yellow warning HDD:
still works
Things are great, I'm testing the new model after training, it seems to even work)
Bro this 5G is freaky https://www.speedtest.net/result/a/11172921915
Yooo, that's dope
It sends more than it receives
Let me train
Almost 1:10 ratio, instead of the usual 10:1 
Yeah.
And let me train please
k gl
Thanks 
And it didn't go well 






i'd rather you just ask here
