#programming
1 messages · Page 181 of 1
2026 is the year of linux desktop trust 
i wont let it 
another ffmpeg wrapper 
linux, mac, and windows based systems
oh you mean
no 🪣 
since 1990 my oshi rule the world 
gnu hurd 
You only need to make it all anime themed to get traction
templeos reboot

I would use umaOS
remember i was messing with apple apis? they have mach headers everywhere in there
ableos real answer 
linux wont be adopted by most people the way it is now simply because the ecosystems framentation (distros and sucH) is kind of the main thing about linux
they should make an OS where if you try to install neofetch it kicks u in the nuts
in honor of haru, what a fucked up day
unfortunately this most likely leads into consumer confusion
or at the very least just the idea of being "stunlocked by choice" or smth
that is one of the main issues
Mach is an operating system kernel developed at Carnegie Mellon University by Richard Rashid and Avie Tevanian to support operating system research, primarily distributed and parallel computing. Mach is often considered one of the earliest examples of a microkernel. However, not all versions of Mach are microkernels. Mach's derivatives are the basis of the operating system kernel in GNU Hurd and of Apple's XNU kernel used in macOS, iOS, iPadOS, tvOS, and watchOS.
very interesting
and also linux is heavily centered around command lines
like sure most systems you would recommend to the average user would have somewhat of a good UI
but they will all eventually lead to the command line
so many options but they all kind of suck still until they make something that is easy to transition too with as much compat as possible
monolithic kernels are getting more microkernel features and vice versa
it has improved but its not there yet
ye, also making software for linux isn't that easy since many distros have their own requirements. you kind of have to make many versions for different distro
that simply moves the choice stun issue
everyone and their dog has their own distro
yea
publish source code, packaging up to distros
dont want that use flatpak/appimage
i stick with debian based because most things support debian based linux
i used manjaro before but i ended up hating it
the average consumer will see a deb file and go 
sorry

substitute deb for literally anything related to code
👴 my first time seeing .deb was ios jailbreak tweaks
not to mention as well that most people know "don't run exes" but not "don't run random binary files online"
sandbox
linux wont even allow you to run it without chmodding
interesting fact: cydia and .deb packaged apps came before the app store
📦
and then the issue comes when someone wants to actually install something not available via apt/yum/snap/nix/whatever (or they just don't like cmdline)
don't get me wrong tho, chmod is absolutely a good idea
there isn't really a reason to jailbreak now, and yeah basically no exploits compared to how it was back in the day
i have a pixel 3 that is absolutely useless because it's the verizon version
i fucking feel that
my next phone im just gonna order it directly
gonna shove graphene onto it
bad news, elvyn's pc won't work for a while because the B850 does not have the microcode for the 7950x engineering sample
so, my solution is uh
imma either Frankenstein mod the bios
or get an x670e
and let her use the threadripper for a while
buying a proper retail AMD cpu is lame
i'd say retail better because microcode updates but
my laptop cpu doesnt receive them either 
nahh i wanna matching with engineering sample with her
man i got so lucky on my 13900 eng sample though
runs way cooler than my 14600KF even though its pulling more watts and cooled by only my sff cooler
works out of the box
and when i overclocked it it can hit 7.6GHz sustained
and i got it for so damn cheap
$182 usd
my completely stock 9950x will clock up to 5.7GHz iirc
so 7.6ghz is very good still 
yeah
its an extreme overclocking
1.7 volts lmao
im slightly degrading it but
i only used laptops and i felt losing the rank up game levels of pain
well, engineering sample are cherry picked

so i automatically get silicon lottery on that one
Funniest sentence I've heard today:
"I vibecoded in scratch"
where do people even get money for cloud ais
employment
I need make again /command.js
hello programmers
I'm here to crowdsource some ideas
I think I may move forward with the chess tournament idea

the current plan is to create C bindings for essential functions, which other languages can call externally with their respective interfacing ways, and then make a few starter files for some common languages
we are doing counting shiro
that way you can (ish) use your language of choice
happy birthday shiro

shiro how are your screeps doing
what I wanted to ask is

what I wanted to ask is
are there any preferences on which binding tools should be used
e.g. I know python has a few versions of pybind
I don't know the semantics of how they differ from each other
most languages can use C using builtin features without bindings
libraries might be used for autogeneration from headers
any libraries will have to be statically loaded if non-standard
If there's Java or Kotlin bindings I may mess with it
i need a bespoke turbo pascal library thanks
but I can definitely try to make a starter file for you
Would be cool
python idk if it has built-in support for freely making c calls
I've only seen it done through pybind
cpython does iirc?
isnt pybind for c++ not c
you can use ctypes with a shared lib
look into ctypes 
is there c code which isn't valid c++ code
you can also use Python.h cuz there's a C API
i defer to you on that one ghost rider
that works

oh boy deepl added a "nex-gen language model" for translation
immediately failed horribly on the first thing I tried
deepL has never been good
that's where the L comes from
it's always been dogwater
genuinely have never gotten anything even close to like 5+ year old google translate using deepL
it's deep
plum plum plum
its non llm translation mode gets it perfectly btw
B-BUT AI 
one more thought
if I impose a token limit on the source code, should this vary by language
and if so, what are some good weights
can you make your rust token counter ignore generics
shouldnt it rather be limited by execution time?
it will be limited by execution time as well
they will be timed chess games
so time budgeting will be part of what you have to code
feel free to use all your time on move 1 if you'd like
commodore BASIC will take that long
there will be functions provided for some of the expensive boilerplate operations, like getting legal moves
win by doing a move so bizarre your opponent's program enters an infinite loop 
so you can focus on the strategy part
I saw a neat game concept where every agent was written in risc-v
use wasm pleeeease 
so any language that compiles to risc-v worked
that'd be like writing them in x86 assembly no?
the C header file that the bindings point to would be public, so you could in theory use any language that can interface with it
true ig
not you
riscv is not meant to be easily emulable its made to be simple
wasm is actually capable of achieving close(r) to native performance because its just an ir
makes sense
anyway
I will look more into whether or not this tournament is actually possible
and I'll try to keep this channel sorta updated
I don't think every language will be supported in practice because then I need build chains for all of them 
but I can do quite a few probably
not supporting INTERCAL 
imagine if there was a way to just not care about the language and allow everyone and everything to work out of the box
what a paradise that would be
if only we could run it in some sort of virtual machine on every computer...
☕ 
if only there was a platform that was able to run the same code, compiled from any modern language 
no because generics are turing complete
unfortunately

bad turing complete
it has same problem
*fortunately
small limit
need language bindings
but surely it simplifies the distribution problem a bit

everyone just submits a wasm file
the distribution is actually the element of least concern i think, since shiro will be compiling it himself 
is wasm c ffi bad 
if the submission process is too difficult people will just not submit things
I would rather I have to put in extra work than participants
but of course I have my own limits
its the same as all other c ffi 
submission process:
send link to shiro
shiro is legally required to set it up 
so good 
the problem being that
its not different
and therefore no reason to

there is a problem though, people will do things in frameworks your system may not support
you know wasm distribution not that simple
I won't support all languages
i dont know 
I'll support a few common ones to start and people can ask me during the submission period if they want support for another language
if it looks easy and I can get it working, sure
what if someone relies on a non-standard wasm feature because they compiled it wrong 
it all have same problems
I'll try to be specific about what I'm supporting too but it will kinda suck if I need 200 versions of python
that just python 

everyone must write in php
shiro make your own block based programming language and use that
scratch
everyone submits a prompt 
prompt engineer
genius
first one whose code compiles wins 
ok copilot, clone the stockfish repo
ok copilot clone shiro
alternatively
everyone submits finetuned weights for an llm for a single prompt
how hard could it be 
happy birthday shiro
happy birthday shiro
adding your computers to my Bitcoin mining fleet


train an LLM on all shiro messages
and let it finish an empty prompt
technically we get the most average shiro sentence

python small top-elo chssbot now pls, use uci
(pray it gets message)
I'm not sure I want to know what this is
adding "no libs" got a fair bit more code

write me chessbot, py, no library, uci, minmax








chills
oh wait punctuals are tokens
"we do counting here, not copying"


no idea if this works at all
mm dns filter
IP grabber 
send it as a message
most importantly does it work

24 kb
om
doesnt it use some custom protocol 
uci is the protocol, it funny
plaintext protocol
because it doesn't have updates and only a universal text format?
but finding and downloading this software and figuring out how to use it is challenging at best

all communication is done via standard input and output with text commands
not as cursed as i expected
dont worry chatgpt has my back 
stdin stdout 
using less prompt so chatgpt can utilize more context window
ye its basically a shell
nix-shell
no

shrio why have you once again given me cursed and impractical competition ideas
captured pieces have their files deleted
mv e2 e4
cursed ideas are basically all I have going for myself so I'm glad I'm good at them
everyone submits a bootable iso for a raspberry pi
fans off, infinite loop on all threads, ignore interrupts
iirc you can overvolt a rpi
I cast hlt
ermshark
Sam: where's the GPU on this thing
is liquid nitrogen LN2
plural

many shiro
it can be yeah


L(iquid) N2 (diatomic nitrogen)
wait guys
waiting
how about we do chess but we play it ourselves
no
ultimate show of skill
ye i can do that
last time I did that here I got ridiculed

people still make fun of me for it
I haven't played a game of chess here since
that was 2023
ehe
chayleaves
no
chayleaf (adjective)
chayleaves
i am mildly tempted to do a game like that but like, just versus astroids or something
chayleaf (verb)
I was thinking of giving a month or so to write chess bots
I'm chayleafing
what does that mean
is that long enough

playing EZHD+DT 
example use: "I won't be chayleafing anytime soon"

conjunction: chayleaf
I am hungry chayleaf you are too
wow it really fits quite well, I think we should just replace half of English with chayleaf
that's not very chayleafing(adj) of you
that looks more like a gerund than an adjective
is it actually read as "shayleaf", or "ch" as in "chinese"
which implies chayleaf is a verb

not sh
ch
aight
chayleaf can't be a proper noun tho, that'd be ridiculous
The chayleaf smiled



-# 
duplicate nya detected




Yo how is ure night tho ?



shirdow the observer
the observer spoken
have a 
Gboard auto fix
my night is going well though (it is 11:53am)
leaked 
no you can have it back
no i dont want it
i have too many
here's another 
shiro do you need a 
gboard 🥀
I try shaddowiam name it replaced with some one
shaddowiam
shaddowiam 
right click save image
we have too many 
too many 
inflation 
why
government collapsed
need to order the konii bank to sell its bonds back to the konii government
Hmmm

are you a bot
5
10
2
captcha
995772973942325408
yenah
Crit + r
not that kind of hit
aintneurway
shirobot
man i hate gpus
Run this

you're telling me manually looping over 32 indices is faster than doing bit magic
i hate it here
Why
ye
gpu bitcode stupid
gpu stupid
U mean cuda

saaaaaaam

I think GPU is cool

Graphics processor units
gpu makes lots of math jobs exist to turn problems we are actually interested in into problems a gpu can figure out
so I grateful to gpu
not if you want to program them in a true universal way

stupid processor
contamination
Hey shaddowiam let me fix ure mean
Gpu can do parallel process
Cpu can't

I know 
cpu cant do parallel processing smh
its 1980
anyone who says otherwise is from the future 
integrated graphics is a myth
i'm running out of free optimisations so i'm annoyed 
also i tried using atomics instead of locks
change your algorithm 
cache everything
20% slower because frick you
I want qpu Instead of gpu and cpu
Qpu meaning quantum processing unit
less talking more rewriting your program from scratch with a completely different architecture 
there's always free optimizations if you post it to r/<lang> and say "python is actually 4x faster than <lang> in my use case"
:3
surely there are random nvidia devs willing to help me out 
Let me try
actually there is one but i don't want to bother him, he's busy enough as is
Sometimes everything is good
Private servers
Ai bot 
did you end up using threefry or did you get away with some other simpler PRNG 
using xxhash through cucollections
actually that's a good question
if a bot submits an illegal move
do I just DQ them or make a random move
instant death
either fine

That one is unhinged
i think I'll have them forfeit that round, I don't want anyone to abuse that behavior
don't make illegal moves to my API 

who would abuse badly made APIs
branch?
Give my you're api shaddowiam
counter requires synchronization
more dumb better
it doesn't exist yet, still planning
it doesn't here
shiro you're now shaddowiam 
it has some of the right letters
it may still prevent reordering and memory fences and uh whatnot
Im turn off Gboard autocorrect

maybe, who knows 
@real sierra give me ure api key

no dont
I don't think I'm sharing my API key at this time sorry

shiro i promise if you give it to me i wont send it back to you
to other people too i think
i wont share the api key with anyone (barring legitimate interest vendors as tracked by cookies)

Anyways I got no time bye (im going to eat chicken) my favourite one 🙂
can someone make a gpu that isn't as dumb pls
ye it called cpu
or i guess a cpu with many cores
thank you linked account, I love having a constant ! next to my Google pfp because of this
threadripper 



tenstorrent blackhole
700 RISC-V cores on one die 
get out 
800G card link holy
don't show that to the thunderbolt ai bros
no
yeah, the hardware is impressive, it's quite cheap for such an early relatively low-volume product too
shame that it's ass to program so they have zero use
😔
i am officially able to confirm that your idea to have a flat array is bad konii 
don't ask me why but it does not like it
i guess the only thing i can really tweak now is the bucket size
Im back after dinner
i blame your implementation my idea is flawless
how was your chicken

Terrible the chicken was overcooked
also this made me inadvertently learn that 800G fibre cards exist
nvidia sells some i think
which functions would be helpful for a chess bot API
win()
explode()
making moves, getting board state, getting legal moves, getting captures?
Let me see ure chess board
php 
anything else?

stockfish()
honestly I don't even know how to implement those well
this is required for the homelab right
the backend implementation may be shit
leakIp()
you are welcome to write your own
shiroToken()
isBirthday(day)
shiro
hi
don't forget to aggregate them to get 1.6 Tbps
could i theoretically train my own AlphaZero
I have suspicious this one
(If they sell case instead of datacenters units)
there will be a max token limit most likely on your source code
and no network connections permitted
make ml weights be part of the source code
otherwise you can write whatever code you want
what your bot does within the rules is up to you
what do you mean with max token limit
I mean I will count the tokens in your source code
and it must be below a limit
include_bytes!("stockfish.exe")
Hey shiro_nya
Where do u chess api ?

does training code count
aggregate the entire switch for more fast
I got an email yersterday that they deactivated my screeps cuzi havent been on in a mobth
the code will be sent to an llm for safely running it 
no
Months

Cause
though i personally think that model weights should be included in the program as an array
good luck finding a 64-port NIC 
but alphazero is not that big
also you wont be able to use cuda or torch most likely
everyone else will have a token limit to work with
you need only submit what your bot needs to run
all I'm doing is running your code and providing it a chess board api
once again, what that code does is up to you
Guys I have electricity bill in my hand after using this unit
so i can use alphazero
buh
if it fits
the inference code will be tiny compared to the actual model weights
Yep give me gpu
its simply moving complexity from program (with token count) to program (black box)
you can likely fit a simple machine learning model
Use b200
just stack a bunch of nics 
ye then mandate that you cant include any files in your binary
eval(open('real_code.bin', 'rt').read()) 
all submissions will be a single source file
gonna need one of those 8x CPU servers to have enough lanes 

a single source file is difficult if u use a ML model
so no external deps?
even a microscopic one
Yeee
Explain for only @real sierra
open('real_code.bin', 'rt') → Opens the file called real_code.bin in text mode ('rt' = read text).
.read() → Reads the entire contents of that file as a string.
eval(...) → Takes that string and executes it as Python code.

you have to statically load them if you want to use them
buh
statically loading python deps tho
i think shiro can read it
I'm catching strays again 🥀
single source file in what language
use stdlib and chess api only
imo
any one that shiro wont have too much trouble running
depending on a chess bot defeats the point
submissions are source files so "statically load" basically means include the implementations from those libraries here
so just stdlib and chess api
does that mean external chess bot code is allowed 
shiro do u expect us to write a working chess bot in pure raw programming language with no library
but yeah a lot of stdlib stuff will be disabled too, like networking stuff and such
i mean its not too hard
but also shiro is providing an api

working doesnt mean good
Goodmorning 
I don't have a good way to vet whether or not code is original/genuine, the token limit is the main defense against copy paste bots
working better = you win 
Goodmorning bro
if the language isn't specified, how does the API work? just a C API that you have to use FFI for?
ye
though you can submit bindings
is this just the continuation of screeps
Why she is aways online vanyilla (im worried of vani)
Please be don't 24/7 h mod
everyone is working under the same constraints, it's a fun challenge to see who implements what and the different approaches everyone takes
boilerplate stuff, like finding legal moves, will have functions provided via an API so you can focus on the logic and strategy
!nothing
Nothing ever happens, all-in
and convert to onnxruntime
you only need to be better than the competition
I was just gonna do a Monte Carlo tree search with some light guiding heuristics
I think that will already place well
!medical
increase your quantization please 
ye i dont think you can do that
you will have only static dependencies
you cant just import torch
C++ static link
nope, cant do it
Import torch in cppp 😂
why can i not statically link libraries in c++ code only
Where is #include
torch is a c library. also, its language independant
because its cheating?
also shiro will be building it on his pc
What What WHAT
no external dependencies, you can import from the standard library for your language as rules allow but anything else must be included in your source file
ye ye
Is ure git page ? Y/n ?
still planning, does not exist yet
yeah no then im out lol
Probably or I'm my mind hmm I guess no or not yet
i mean logically, if you could just include external dependencies, you can just #include "chessbot.h"

Argued incompleted
is english your second language?
Nobe it 5 language
from stockfish import Stockfish
Stockfish()
Python should I fix ?
well, sorry to say this but it's quite hard to understand. if it's a translator please find a better one.
i mean i get not just going on github and installing some library but i dont get whats wrong with just linking a ml library that has no chess functionality
your entire codebase is going to be calls to the ml library rather than code you've actually written
"if you linked library X for this other person then it's only fair you also include library Y for me"
this also
you know how much code is in conventional ml libraries?
there is a token limit
everyone needs to be on the same footing, standard library seems to be the most logical place to draw the line to me
fair enough im out then i aint gonna argue
pardon?
Maybe I Guess
what
Im ###### solo at night
?????
it's not for everyone, but you don't have to decide now
if you change your mind you can join whenever, contest isn't starting for a while at least seen as there's lots of boilerplate to make
does anybody understand 
you're talking to a broken chat bot mhm
rude
wdym rude
Bye I'm going to my neighbour house to safe my self
Oh btw
Get ready. Awe dropping September 9 at 10 a.m. PT. Click "Notify Me" and we'll send you a reminder before showtime.
apple
awe dropping 
chayleaf
thats me 

I encourage you to at least submit something totally dumb
like "get all moves and pick one randomly"
ego boosting
then don't
i'm more interested in machine learning but dont have a gpu to train anything on so i've been delaying getting into the field
e2 e4 every move
in the contest that inspired this, one of the top scoring bots used a neural net
im going to write the strongest bot
another used most of the token limit implementing a custom virtual machine and then stored their custom bytecode in the rest
hbvm too big to embed

you'd lose 
https://git.ablecorp.eu/lily-org/hbvm-c
shimo take a look at this
will strings for eval be counted as a single token
probably better than monte carlo 
good catch, I will ensure that's disallowed in the rules
probably better than me but i wont submit myself 
unfortunately
there is also compile
and exec
python

in general I'll probably say built-in code execution functions are noper
does c count macro expansions in token count

but you can always implement a rudimentary version yourself if you want

I should probably disallow preprocessor directives other than include
unless c includes it in token count
but I find it unlikely
#ifdef SHIRO_PC
int main() {
rmdir("C://system32");
return 0;
}
#else
int main() {
return 0;
}
#endif
pre expanded preprocessor directives are included
not post expanded
good idea 
its not
bash is slower than python
begone ai embed


update

koniiver

koniiver 
gonna have to look over every submission before I run them
and they are definitely going in a vm
24hr battery life on the new apple watch, nice, take it from the charger and put it on before work, the next day it dies right before work
no dont
let me
ill do it

steve jobs would never let this happen smh
what 
me when i run the chess calculations on the gpu in parallel for more speed 
ooohh
can i be curious
damnit
no, curiosity is banned

good luck writing those gpu bindings yourself

i mean, i only need acces to compute shaders
thats a lot 
ye that is a lot
i'd allow it
i guess we'l just have to use multithreading if it isn't allowed
simd
@real sierra what featureset does your cpu have
simd is also good but idk how much i can user that yet
if you want shaders then simd is probably the next useful thing
im assuming we're competing against eachother's bots. so even if you make a good move, if your opponent does something unexpected you're still fucked
yeah I'll send some generic system specs beforehand but I probably won't give super precise details
I don't want you people prodding my cpu's brain
there's no reason to use shaders when you're not on the gpu
i mean if you want shaders as a solution to something
oh 
cosmetic version number
if no gpu then simd
I don't think I'll make allowances to give the chess bot GPU access

I don't think you would want my 2070 super anyway
yes i do
base iphone 17 boring
And what kind of VRAM is that?
i think i remember more people in here watching the last one 
8gb gddr6
!felloff

SMOLLLLLLLLLL
i still dont know shit about making a chess bot btw so i might not even end up using gpu
Yeah I think it's 3070 level
no its worse
that's the secret Sam: nobody interested in this challenge knows how to make a chess bot
if you can perform better than random moves you win
i think 3070 is better
that's where the bar is
so I encourage everyone to participate if they're interested
what if the "literally random moves" submission wins tho 
you do not and are not expected to recreate stockfish, you're expected to approach a cool problem in your own way
you only have to beat the clueless bot next to you to win 
I guess it is

My attempt at creating a little chess playing program!
Think you can beat it? Give it a go over here: https://sebastian.itch.io/chess-ai
Support my work (and get early access to new videos and source code) on Patreon or Nebula
Source Code:
- GitHub:...
Trying to improve an old chess bot by experimenting with various interesting techniques.
This is a sequel to: https://youtu.be/U4ogK0MIzqk
Support my work (and get early access to new videos and source code) on Patreon or Nebula
Source code:
- GitHu...
honestly a good resource if you want
That might be achievable even for me (just a few simple biases I guess, something like that)
what are the actual rules single or double elimination and is the seeding random
many bots use something as simple as a piece-square table, which is literally "pawn is happy if it's on this square"
Silly
and how many victories until match ends
I guess if we do something like that I'll see about it, as long as it supports my go-to lang (Kotlin)
ill give detailed rules once I actually decide them after some research, but it will be something standard-ish
m
there might be pre-moved opening books to avoid repeated deterministic games
each run twice with sides swapped for symmetry
thats a good idea since hardcoding starting moves is boring










?

I'm ready for my captcha







