#programming
1 messages · Page 202 of 1
it's actually not all sitters but still that was a lotta packages
aint no way
i preferred gh on site wayyy back in the day when i fought with git all the time
Ermagash finally! Coming from a country where 10mb is the norm, this blows my mind
lazygit is nice
😅 dont know why the image is so huge sorry
yoo
nice
that must feel incredible
in case you forgor
yeah and you press ? at any point
interactive menu which is context aware with all commands
you can click on the branch/worktree if u want or just hit 1/2/3/4 to select an area w/e
👋
i need to disappear to get some form of sleep as well or i will pass into the void
gn yall now i wont be clogging ur chat
if you don't want CLI, you can use something like github desktop for gitkraken. there are bunch of desktop git apps
Goodmorning 
morbing
@real sierra I have both good news and bad news about the latest fuzzing run
Score of rust-chess-bot (release) (a) vs rust-chess-bot (release) (b): 5590 - 5654 - 88756 [0.500]
100000 of 100000 games finished.
Player: rust-chess-bot (release) (a)
"Draw by 3-fold repetition": 81961
"Draw by fifty moves rule": 712
"Draw by insufficient mating material": 4153
"Draw by stalemate": 1930
"Loss: Black mates": 2847
"Loss: White mates": 2807
"Win: Black makes an illegal move: e8c8": 1
"Win: Black makes an illegal move: e8g8": 1
"Win: Black mates": 2738
"Win: White mates": 2850
on one hand, it did 99,998 games without crashing
on the other hand, black got unlucky and played two illegal moves 😭
both of which seem to potentially be trying to castle
(sadly I do not have the pgn of either game, just know that they exist)
also these perft results don't inspire confidence
0: 1 (done in 60ns)
1: 20 (done in 7.45µs)
2: 400 (done in 28.581µs)
3: 8903 (done in 465.949µs)
4: 197282 (done in 9.812143ms)
5: 4865460 (done in 238.386248ms)
6: 119047203 (done in 5.820904955s)
7: 3195714202 (done in 154.238689357s)
(should be)
thats based
dead chat 
audio was terrible
who THE FUCK at apple decided that mixing in iphone apps into the fucking app list was a good idea
surely there is a way to turn that off 
Hmmn
i found it but would've been nice to have them still be searchable
MacBook apps
AreSoo synchronise
there is this thing
yeah i want the app list to be the list of apps installed on my mac since they removed the launchpad or whatever it was called
for it to be like that
fair
I will try making any charter in graph
this icon is so stupid 
how it used to look, for reference
apple massively falling off
how does that first icon even make any sense 
well you see, disk utility
and a disk is a shape
you know what else is a shape? a hexagon
do you know what a hexagon is used for? bolts
and do you know what bolts are tightened with? wrenches
and disk utility is basicly somewhat related to repairing liek how you repair stuff with a wrench

repairing apple branded hexagons 
take to genius bar
genius bar recycle your device and sell you new one for discount disguised as repair
only apple technician can help you repair Apple® iHexagon®
I am more than tempted to try and write a tiny little pxls backend in rust or something
because there seem to be serveral questionable choices with the java backend idk
things that need to scale like that should be the first to riir
727,727 does need to be in an entirely different system in case someone decided to start meming.
-# early send due to mobile
are the coordinates 10 pixels below the center of the osu logo on all canvases
ngl the biggest thing that pains me is the postgres database write on every pixel placed
WHAT
WHY

Surely it was for "an accurate log"
even then you can batch
that should be handled separately
or store userid with the pixel data
like all this needs is a a bunch of connection handlers, a queue of incoming placments and a worker on that queue that writes occasionally
Or even just an in ram database to go fast and batch from there if you need to
yeah you see
it does have an in memory database as well
just every write instantly goes into the db
does this at least not block

because it's a lambda, maybe it's in a separate thread? maybe...? please....?
I'm genuinely so suprised noone has rewritten this
because like this is a trivial thing to optimize the hell out of surely???
Instead they have undercooked frontends
backends too apparently
I can't talk, I'm worse at frontends
Just make it modular and offer theme support, it is not that hard
all I know is the sheer amount of "the backend died" this canvas were far too high
like a pxls instance should not be dying with 1000 active players
First day there was no audio slider at all, hardcoded 100% volume
the problem with these kinds of projects is if an actual frontend engineer works on them people will be like "le <frontend framework name> is problem you're at fault for bloated web we don't need your fancy crap"
"go die" or something
htmlx, the ultimate no u
do you mean htmx or is that something i'm not awware of
please don't use htmx for a pixel canvas frontend
that would be like an ACTUALLY bad ux
Now that I'm thinking about it, the canvas object would die on contact
just use a massive html table
3fps 
locked out due to out of ram 
ngl you could even do some nice stuff with refunds if you made it all into a queue
tbh wouldn't be a problem if the sound effects wouldn't have been literal 0dB square waves/pwm at like -2 LUFS
instant deafness
The cursed solution I'm considering is live patching an image file. I'd not recommend that tho.
I am part of the problem yep
like have each placement be ```rust
struct PlacePixel {
color: u8,
command_at: Instant,
}
and then when the queue finally gets processed if `pixel.command_at > incoming.command_at` just refund the user ig
ehhhh it is kind of how canvas works
except the image file is wrapped with an api
ever right click -> save image on a <canvas>?
Nope, haven't tried that before
some websites just override right click or disable pointer events on canvases
but that is the default behavior for them
this probably introduces some shenanegans with clientside prediction but w/e
surely with a competent architecture you're only gonna have to be waiting for ping and barely any actual processing time
you're thinking of optimistic updates, any reason to do them in the first place?
or am i misunderstanding
why have command_at if they are in order in the queue
oh true
and it would not be a "refund" if it fails
nah
because as far as the frontend is concerned someone changed the colour after you did
I think optimistic can be fine here
because worst case it gets rejected
actually no
I guess
but on the other hand if it does get changed surely that wouldn't matter
I should probably yoink the diagram for this one ngl
The current board has green (succeed) and red (failure) response outlines
I mean I'm just gonna try and make something compatible with the default frontend if I do
or I guess I could chuck together a budget frontend so there's less silly rules
to me it seems like the actual <canvas> part is the hardest on the frontend side
but i do frontend so
even then is it not just a canvas that paints squares and can be zoomed
That and any overlays people have
yeah but then you have zooming and panning and overlays
if you want to do it correctly at least
ok but surely overlays aren't much more complicated?
you can get away with moving and zooming with css but it would not be a good ux
also can we get msaa for the frontend pls okthxbai
probably not if you have the "correct" way set up?
because if it's just a 1000x1000 <canvas> you can't do overlays
because they require higher res
?
overlays are more dense to have numbers or symbols
right, that's what i mean by "correct"
1000x1000 is the res of the canvas, the <canvas> should be the res of the viewport
confusing stuff why can't they be named differently 
ok so basically <canvas> is an html element which acts the same as <img> but you can render to it from js
Hmmmmmmmmmmm mMMmMMMm
I think some
Mess up with Lenv
ye, am aware
you should not directly mirror the state of the canvas to that <canvas>, it being x1000px and all
rather that <canvas> should be the size of the browser viewport
and you scale and pan in the rendering code
I don't think I can even be accurate down to a pixel with my mouse lol
don't get me wrong I should be able to but something somewhere disagrees sometimes
i have my mouse at 3000 dpi on 1080p
every time someone uses my pc they are shocked
or was it 3200
you get the idea
no problems hovering over a single pixel
lmao what
I was checking my mouse settings to double check my dpi
100ms delay on my bhops
yes please

accidental jumping is an actual issue though
ye I run 800dpi
I belive I'm at 800dpi when I'm not using trackpad
especially in a game like cs where everyone feels obligated to bind their scroll wheel to jump because they're the coolest person on earth who definitely for sure knows what air strafing is and isn't doing it because they feel like they have to because their favorite pro player does the same
but then 10 sens in windows
I mean
I do bhop in games where it does something
cs is not one of those games for me
Chat i might actually be genuinely cooked
I applied for a job position in Software engineering where they do java development
either way you start scrolling before you hit the ground so that's not that big of a deal i think
ig
fair enough
I just like not having random points that throw latency into what I'm doing
And the company had asked me what my career goals are
And i told them ML. They said fair enough and now i have an interview with the company's CTO and director of R&D ML research
god bless open fortress
actually such a fun game at lan
I might be genuinely extremely cooked
though for some reason its really hard to explain bhopping to people who've never done it before
no but that's cool though, if they like you enough you're in
that's not a random recruiter you're talking to
nah, you're fine probably, it just means they're interested 
you'd be cooked if they just put you into some random Java position
as in, web dev or maintaining some legacy non-ML-related app
So it's a good sign?
definitely
Very
important people are taking their time to talk to you
Why tf would the head of r&d and cto contact me tho 😭😭
Like isn't interviews supposed to be with HR
beacuse they're interested ??
no, having technical people on interviews is normal
having the CTO isn't but either that's their policy or something or they're genuinely that interested
Shadow, as a person of the same type as you. Your not giving yourself enough credit
I saw they have a research org on HuggingFace
So is it likely not for a java Position im applying?
they don't necessarily have to give you an offer for the position you applied for if they have one open elsewhere (as long as you agree, obviously)
assuming you already talked with them once
You probably applied to a java pos, but ended up at an ML pos (Same as above)
I mean I'm fine doing java
It is your choice, I'd recommend the ML position as it is what your interest is in
you have actual (not job) experience in ml and (probably) have actual stuff you've done to show them if you're concerned about proving your ability
as for the interview itself when i see important people like that on interviews it's usually half interview half vibe check in my experience
Wym half vibe check
I do have 2y job experience in ML at google
seeing if you actually want to do ml, how good of a fit you are to the team and generally fitting company's goals, and how much they personally like you
what i mean is experience outside of job experience
not that job experience isn't real experience or that it doesn't matter or you don't have it
I see
e.g. in SV it's common to hire engineers based almost entirely on how much they like you
Bruh no way
i understand you're not applying for a SV startup but that still kind of happens
if ml r&d is there then probably?
I'm scared they'll pull some phd level question
you still have the java pos
tbh even if they do, they might just actually be interested in how you approach it at that stage
The original position i applied for is Java for enterprise resource planning
Wym
like checking how you approach problem solving something you've never seen before, to make sure you'd actually be able to solve problems you get given in the workplace
Oh so ask an impossible question basically?
maybe not strictly impossible, just one you've never seen before // involving a bit of thinking before giving an answer
Oh yeah no i mean them for example pushing some phd university exam Level question
Abt some intristic mathematical detail
does the position they're considering you for involve those types of questions
They only have a position open for java dev and accountant
At least externally
right but, what kinds of ml they're working on, do they involve that
They work on llms that are put on enterprise resource planning software
since you said they have a hf org and stuff, from your assessment
Question answering type assistant
you should be good then i think?
you shouldn't stress anyway
but
i have no clue what they'll be asking you
but i know you have experience with llms
you'll still be able to fall back to the java pos probably if they don't end up seeing you fit for whatever ml pos they're considering you for
if you're actually fit for the java pos at least
lol
no i'm not 
gosh i will never be used to that profile pic


I tapped on the wrong channel mb
i almost got lost in these wild and unfamiliar lands

honestly programming is kinda osu channel too
#osu got split into #programming and #gaming

just 89
new number

wait actually


i like tech (maps)
bad


sliderbreaks sliderbreaks sliderbreaks :mhm:
you play them for fun not for combo 
I only play for pp and ii
I don't need fun
tech maps are great, just not in my skillset, yet
been trying it for future tournaments
I just got finished playing a few songs in beat saber 

-# never is not charted yet 
chart it yourself 
Don't know enough on how to chart for any game to chart that
☝️
(i have no idea how hard beatsaber charting is)
beatsaber mentioned
Editor is a 3d grid that you scroll through mhm
Not as much as playing the map while testing
testing charts you make? why would you do that?? 

I only tried it before they introduced streams so the UI is probably a lot more different than it used to be.
I still struggle a bit when they are used
streams being those like super thin notes?
Yeah, I always go early on them
so half of them don't get sliced
holds are not as much of a pain but hard to properly track sometimes.
Yeah those
you don't need to follow those
they're just a "hey move your saber like this otherwise you'll get jumpscared after the break"
I don't really play acc, I just play casual. Mostly on the starting maps but I have at least played the neuro songs.
i have not seen beatsaber gameplay in a long while, holds? streams? i thought it was just cubes and the barriers you avoid 
It has slightly more in it's toolkit now
I kinda don't play a lot but https://beatleader.com/u/76561198093022590
I still can't quite deal with every expert+ yet, but I'm getting there in between play sessions (which are rare)
it looks kinda goofy because my first ranked play was 148pp lol
but like if you want there's replays on there
Don't have an account on any of the sites. So 0pp for me.
I don't think I've even played any ranked maps now that I think about it.
i mean my account got made automaticalled because I have both the scoresaber & beatleader mods
Either way I wouldn't have an account because I share the headset and don't have beat saber set up on PC atm.
No point if I play rarely anyways.
it should have been the easiest option to throw a web ui together in theory
but i blinked and now we’re passing state from component instantiations to other components
and they call actual frontend frameworks bad and overly complicated
i don’t understand state in gradio no matter how much i try. i feel like it’d be easier to forget about it and just write a flask server and throw together an actual frontend
ok maybe i’m starting to get it now, there are like 4 ways and only one of them is sane
still too convoluted of a workflow
man i hate python
i read the docs, think i get it, unminimize code editor, head empty
ok so all components own their own state and can pass that state to other components
why
castles, huh
how do I even find a bug like this
when you control Z 200 steps back to copy some old code and accidentally press the wrong key so you cant go back to the future 
it's clearly not every castling situation
I would start with that thing I mentioned later about the perft values
only about 1 in 50000 games encounter it
no idea what a perft value is
because if those don't match you're doing move generation wrong
but if it's generating an extra castle move that would basically explain all of it away I think
^
for clarity that was generated with
fn main() {
let mut board = chess_api::Board::get();
for depth in 0..=10 {
let start = std::time::Instant::now();
let moves = perft(&mut board, depth);
println!("{depth}: {moves} (done in {:?})", start.elapsed());
}
println!("done");
unsafe { std::ptr::null_mut::<u64>().write_volatile(1) }
}
fn perft(board: &mut chess_api::Board, depth: usize) -> u64 {
let mut nodes = 0;
if depth == 0 {
return 1;
}
let moves = board.get_legal_moves();
for r#move in moves.as_slice() {
board.make_move(r#move.to_owned().into());
nodes += perft(board, depth - 1);
board.undo_move();
}
nodes
}
except I got too lazy to let it run to depth 10
btw
not super double checked for typos but should be fine
the perft page gives a few other starting positions, can you find any where the bad move occurs in the first depth
that would be few enough for me to comb through manually
sadly probably about to head away from my computer soon so can't
evening

64 bit address space
on the web, a 64-bit memory is limited to 16 gigabytes
how will i run firefox
running firefox in firefox 
Shiro so intelligent


builtin strings is huge for wasm 3.0
does this mean no more __wbindgen_malloc for passing strings
Most builtins should be simple and do little work outside of calling into the JS functionality to do the operation. The one exception is for operations that convert between a JS primitive and a Wasm primitive, such as between JS strings/arrays/linear memory. In this case, the builtin may need some non-trivial code to perform the operation and it's still expected that the operation is just semantically copying information and not substantially transforming it into a new interpretation.
this feels like it should answer my question but it does not. i might be stupid
who knows how they do it but they do it
i understand it like this, wasm can get refs to strings owned by js and can pass that ref to built in string functions
but the js host still owns the string?
https://github.com/WebAssembly/js-string-builtins/blob/main/proposals/js-string-builtins/Overview.md
yeah that's whati'm reading
(module
(global (import "strings" "my string constant") (ref extern))
(export "constant" (global 0))
)
let instance = WebAssembly.instantiate(bytes, {importedStringConstants: "strings"});
// The global is automatically populated with the string constant
assertEq(instance.exports.constant.value, "my string constant");

my initial idea was transform all pieces into code which has some piece id, its available directions to move, and its coordinates.
so i could move piece with 1 operation by &ing it with the shift and checking if its not 0 with some boundary mask, so its 2 operations to move the piece and check if its out of bounds.
now its useless since we have the api.. and idk if it was even a good idea.
but focusing entirely on strategy seems more fun anyway
what I really want, what I raelly really want, is for wasm to have its native api for interacting with the dom and the browser
so I dont have to go through javascript
bitwise operations like this are how the API manipulates the board internally
there are 12 bitmasks, one for each piece-color combination, with bits set for squares containing that piece
so it was a good idea after all 
moving a piece is just an xor with its origin and destination squares, getting all pseudo legal moves is just bit shifting and masking the boards a few times
now what about this
i dont ant to work in vacuum 
it's a very ubiquitous way of representing these boards, called bitboards
not until the contest starts officially
the only people with early access are those making language bindings
I think i sent the header file a little while ago
yeah and i cant find it for some reason
i wonder if "file" filter search actually doesnt include real files like .h and only checks images and other media
I'm on phone right now so I don't have it downloaded
i'll ask again, what langs are bindings being made for outside of bred making ones for rust and you making python ones
oh sorry I missed the first time you asked
scrub mentioned making C# bindings, haven't heard updates on that
eh i'm just curious don't wanna be pushy
chayleaf wanted to make Haskell bindings but is slightly sidetracked
what are those bindings? language support?
it should be really easy to make c# ones
language support
allows people to make calls to the C API in their language of choice
wait but
this means anyone who wants to make a bot must have his language supported
yes, you can't write bots in every language
that's why they're being made 
it has to be a supported language
that's why people are adding support for the languages they want to use
should be
lmao
i was confused for a second why your messages swapped order
im solo dev and making bot my self ore some times ai
You mean the "Hi Shiro~" and "I'm here now"
no i mean the images
100 reposted the image in new format and it appeared being after shiro's message
"tuh bluetuh device is ehh connekteed sucksesfooly"
Shiro~ I have some code:
this is indeed some code
is this from the doodling thing i saw the other day
Yes, that is the model training code.
Yet my models for some reason... Don#t perform well
yes 
How many images should I use for the model? (I can go to like 400 only)
holy clean
I can in theory use any images for the model
i'm not familiar enough with this kind of machine learning to give a useful answer
Same, but I'm also not gonna judge you. Anything would be a interesting idea
it should probably maybe hopefully be possible
this uniform buffer shit took so long to implement for independant objects
is book = perfect move?
book = opening move
basically it's a move made according to established chess theory
there's no evaluation or search involved
just "i respond with this move to that one"
so theoretically its supposed to be very good but not guaranteed perfect
otherwise whats the point of doing it if this move is bad
whey wouldnt put a trash move in their book right it makes no sense
chess openings have been studied for like 100 years now
so there's a bunch of very specific openings that are commonly played
all the common ones have names at this point
generally though, they are all quite good
which is "best" is a matter of opinion still
i broke it 
tutel stand
very broken

i think my bot can only spot simple checkmates, at least on this time scale
python 3.14 already exists no?
Pre
Python Release Python 3.14.0a4 | Python.org https://share.google/fwBbnjsKbxgAe2Qy9
Bye I'm going to sleep 🙂 👋
deferred evaluation of annotations
I've never seen this Birthday tag before lol. Happy Birthday to you!
rest seems useless tho 
The free-threaded build of Python is now supported and no longer experimental
no thats pretty good

ye maybe useful for embedding dsls
have you tried playing against your bot @real sierra
-# No because the bot would win
I can't get it off
I couldn't even beat the old version
HUH
for starters I'm bad at chess, but I also am just not capable of playing a full game in 1 minute
i think you are supposed to play longer time
so you can analyze and see when it makes weird stuff
its going to move fast anyway so doesnt matter its basically gives time to you only
I would actually be interested to see how my bot fares against a more skilled opponent
depending on the opponent's move, the amount of my move tree that can be reused varies greatly
against a random opponent the odds are rather low that their response is something I anticipated
but against an opponent that plays similarly to my bot, it can probably do a pretty good job
I considered doing that as a benchmark actually, stockfish has 20 levels so it'd be a nice standardized test
but then I have to install stockfish which is annoying
bwa
https://youtu.be/vFz1HcFEL44
I've been working on a voice to voice bot, really fun side project, having some trouble on latency and accuracy of ASR, might be just because i'm naively using whisper or something but i bet I could keep messing around with the wav data to more accurately detect periods of silence from within the data payloads SoX spits out. Also anything faster than Whisper-small is like almost useless for accuracy, so as what I gets longer the fire and forget approach starts breaking down and increasing latency on the response.
As you can see the actual voice generation coming from XTTS is super quick due to me shotgunning it with sentences, i can't really do the same on ASR though as it's not possible for me to halt running processes once it's passed on to the GPU, and it's queuing up, leading to a 1-2 second delay before LLM invocation
what sort of game can i make in python using MU that isnt text base. Ive seen my friend make a simple clicker game using MU and i want to get started on something myself
With text only you're kind of stuck, could try something like Drug Wars maybe
If you can python you can probably JS with a little effort, then you get browser access and the world's your oyster
i don't think this is good advice, i say just go for a python gui/game lib
Entirely depends on your end goal really
If you just want it for learning to code, then absolutely stick with it
if you want to learn how to make games start getting into graphics early
well,
game in python
isn't text base
i don't think "learn js" answers any of that
Yeah true
I was just kinda saying game options are limited if you work with something that's got no real graphical support
C and C++ awaits!
lol
cc. @olive sable

baby's first python
an editor i think like idle
like a nice IDE
@rigid snow I presume you are just starting out, yeah? What is your main goal, what are you trying to get to with learning?
wrong ping?
ye lol im doing computer science GCSE and this frend did make a GUI looking clicker game
i am NOT just starting out 
He made it for fun and I also want to make a game that would run in python but isnt a clicker bc he already made one
ig i could make a gambling game
Toughy for sure ..
i mean, if you want to make graphically intensive games you'd jsut go for openGL in python, if you want more visual novewl style go for renpy
mint is learning to code
surely there are better graphics libraries for python than just raw openGL 
like... pygame 
ykw i think ill stick to these mini MU codng challenges my teacher set for me for now ig
just dont use python
so true
there is this fishing game i saw done through text
maybe you could replicate that but it's also kind of a clicker
if you use a ton of libraries thats kinda like a game engine but you need to provide all the glue to stick into together into something meaningful
i will never go completely low-level again. doing that in c++ was a pain 
look i did hello world in js 
I do not need this information in life
mint is already creating horrors
they grow up so fast
obligatory: https://github.com/denysdovhan/wtfjs
my beloved
I tried writing a game from scratch in java 
I had this super cool wrapper class I made that would automatically sync objects between clients with no added effort (it was multplayer)
we got NaN 2 
it should be a number that has banned combination of bits
NaN just means the float is busted
i'm so tired of hearing of that as an argument to anything
it is still a number type
JS is very fun, I actually like it more than typescript but i think that was because i was forced to use it so long before TS existed all the mental load is like, embdedded in my head
how many unique nans are there in 32 bit float
Imagine using JS before ES6 btw, you can't
couldn't be me
What's ES6?
ecmascript 2015
imagine having no async
a lot of the codebase i worked on was pre-prototypes too so we were doing just raw factory functions
what do you guys think about lua
I actually love lua, but specifically only for scripting
lua is pog
and pico 8 flavour
lua is python if it was good in my eyes
no need to declare variables
local 
i've never really had the opportunity to make something serious in lua just toy things
one does not simply avoid local
but i had a lot of fun with it
yeah one local and then do anything with it
lua is good for when you want a simple scripting language
like for ui
or easily sandboxed mods
i would much rather have a wasm engine nowadays
https://www.lexaloffle.com/pico-8.php i'm sure most of you guys know of this but all the games are in lua here and the source code for each game is open, so you have a mega repo of lua examples if you are learning
hell, i'd argue it's somewhat decent as a build-script language
totally
way better than the hell that is gradle, at least
gradle is evil, but it's better than ruby 🤐
i've never had to encounter ruby, thankfully
it's syntax is just awful imo, i've mainly used it in cocoapods stuff as part of build flows for things like cordova
apple developer spotted
speaking of awful syntax
Kill me now
I am so annoyed with python dear god
I tend to defer to nodejs for similar cases just because it's cozy, and when i REALLY need python i'll just spool up a server and talk through APIs
I know it's not efficient, i just hate it
tutelarray is working
based
im going to bed now
gn 
see ya man
this will multiply my pay by 25
he pays you? 
yes, the full 0.00249 cents per ad
"you guys are getting paid?"
gradle
gradle good 
android ruby change my mind
android ruby
i cant be bothered, change it yourself
i know
I love that picture
happens
competing with chatGPT at this point
ur a margin of error 
bwaa
the margin of error on missing people wouldn't notice 1 shiro would they? 

more like gaydle
hahahaha
gay tooling = good
this is why i prefer gayven
be nice to each other
where is this running
maven can burn too, i havent seen a single good piece of java/python build tooling
a lot of the really low level functions have been written so now i can chain them together to do higher level stuff. development shouldgo quicker now
my gpu
chatgpt clears
Officially better at math
because the languages weren't designed with build tooling in mind
skill issue
just use c to build everything
nob.h my beloved
nah
Have any of you guys worked with any nice speech recognition AI models besides Whisper? I am in need of fast options i could run locally, and huggingface is very broad, nobody shows latency
i forgot how it was named but there was a pretty damn good local model since forever
lemme look
(begging it's not whisper)
also if we're talking about languages that lack in terms of build tools
c/c++
i love those god damn sons of bitches but why is building with them such a pain sometimes
call me crazy but i'd rather cmake than gradle
I will eval, but it can't be worse than whisper-base, hopefully
it's older than whisper i thinkl
C++ lacks a bit in everything tho. the very fact we have to use header files shows it hasn't cooked enough yet
gradle 
shiro you're a java dev how do you cope
maven
c++ is just c if it went a different direction and kept adding feature after feature after feature
good c++ is good but bad c++ will make you wet yourself in your pants and at your eyes
inb4 rust
C was great and they ruined it
it's this and it is for sure worse than whisper, models per language are 50mb only tho. it's more for embedded/mobile i think
https://alphacephei.com/vosk/
all other c++ is garbage


c++ is the python of systems programming
I hate your c++ enough for both of us
Architecture design
you haven't even seen my c++ :(
i'll give it a try! if it's fast, and mostly accurate on clean audio, and doesn't try interpretting fan noise as '*fart*' 50 times (yes this happened i have a screenshot) then it should be good
sometimes i like to write my c++ like i write my c, except i have vtables at the ready for when the moment calls for it
c is great imo
i hesitate to call c++ anything above manageable
and hating on
https://imgur.com/a/fqDAD8O i have no embed perms so an imgur will do
sorry for your loss
nice embed
tf did you do?
blame imgur not me
i can only imagine it was digital terrorism
why is imgur hating on you both
That site was always really weird tbh

temu reddit
me irl -> 

me irl -> 🐑
me irl -> 
i have not admitted to anything even remotely close to chay
me not irl -> 
"admitted to"?
ahaha
not "I have not done anything [...]" ?
international man of mystery

keeps you guessing
suspect
You have unlocked new role
woah
even Tatsu is suspicious
i am being investigated it's over
it's so joever
oh maan pastebin is so screwed these days
not pastebin!! how else are they supposed to be able to use computercraft :(
how much for the ban to take effect
you used to actually want to look at other people's pastes, these days it's 99% crypto scams
I say you, i mean me

i get crypto scams delivered straight to my email 
gotta have the emojis there like a shiny keychain
jiggle that thang
⭐⭐⭐ Instant Attention Method ⭐⭐⭐
I don't intrinsically hate crypto, but dear god is the degeneracy pervasive
how dare they drag the holy java's name through the mud
I'm declaring war on crypto scams
script to auto report Pastebin entries
the batman except for java pastebin cryptoscams
jatman
you can just report every single one, 0.01% false positive rate
So those new meta smart display glasses look really cool
But
They are the last company I would trust with a camera on my face 24/7
In fact I'm not sure I'd trust any closed source product with that but most definitely not Metas
meta? more like beta
after that live demo flop? no way
I wonder if there is an appetite for transformer designed to detect crypto scams, i genuinely might train that using these as training data
I'm willing to give live demo flops the benefit of the doubt since it shows they were confident enough in the product to give a real showcase even if it goes wrong
Yeah i respect the balls to show off essentially near prototype tech like that live
nothing is flawless, especially with AI
The neural band thingy is also really cool
Yeah that is actually insane
If they sold those separately I could see myself getting one for other devices
Hands free phone operation with one would be pretty nice
All i can think is how that data could make hand prosthetics better
like you get electromuscular data from the wrists of 1M people you can make incredible hand prosthetics probably
I think as far as smart glasses are confirmed Brilliant Labs's glasses with the Meta Neural Band would be the only viable combo right now for my tastes
Nice input device + open source software
But the value proposition isn't there yet imo
mojo lens 
Great idea but I'm scared of putting shit in my eye 
people with contacts do it everyday, I'm not that worried lol
xddd
Not because I think its dangerous
Just because the idea of jamming it in my eye freaks me out
Just looking at paradigms these days, once the AI wave chills it'll probably be robotics then BCI stuff i reckon, both areas benefit incredibly from the shift to transformer models, i reckon. Definitely robotics, if you look at what happened with Boston Dynamics after they swapped from actuators and normal neural networks to transformers and servos their stuff is soooo much better, and it was incredible before
I would be so confused by the lens tho
I would instinctively try to look at stuff displayed
and then be bamboozled when the screen moved with my eye
Have you ever had floaters in your eye when looking at the sky
that but an ad for monster
looking around and the HUD stays in the same spot in your field of vision is nuts
actual video game
maaan imagine them showing ads in your peripheral vision hoooly
this video does not show up in my youtube studio for some reason
OH it's treated as a shorts
why
Damn that's cool af, i am so sad they got rid of that on the new macbooks
i have an M3 and had one of those before, feels a bit naked
but i loooove little displays like that oh man
was the m2 the last 13"
https://nl.nothing.tech/products/phone-3?Colour=White&Capacity=12%2B256GB this phone has a little programmable addressable LED matrix that is so fun to make things for, coolest phone ever
only 25x25 but you can get creative with it, 120hz for some reason
also NL doxxed
I kinda worry my bot has peaked 
idk how I could improve it any with the tokens I have left
I think it'll actually be over the token count once chayleaf fixes the counter
in which case it's truly gg
Lens moves with the eye?
I would have a nothing phone if I wasn't so in love with the Pixel feature set
Whaaat
I mean yeah, it's a contact lens
it's attached to your eye
share bot i am new
well not attached but
They can make it moving the hud dynamically
Yeah fair, i am just bored with glass rectangles
I am neither talking about bots nor tokens in the sense you hope
Öet's talk coding
Lets say i have my hp down bottom, i move eye there and lens moves it to the center
So what are we talking?
So i am staring directly at it now
I'm planning a chess bot contest soon where people will make a chess bot and then they'll all be put in a tournament to see which is best 
but to add to the challenge (and stop people submitting stockfish) there is a maximum source code token count of 1024 tokens
Whats the issue with token counter rn?
very much inspired by that one!
there are some damning situations where it doesn't count token sit should, and it doesn't support as many languages as is desired
I won't probe on your angle beyond this, but heuristic or mini-neural?
MCTS
Shiro~ have you gone to "the toll free amongus car vc" VC before?
yes I have lol
Me, scrolling down for the first time
I forgot discord lets chatrooms get that big now
How's there usually?
Yeah, I just listen to the songs rn. No one is talking meanwhile
if I do somehow have tokens left after the token counter upgrade, the next thing to add is transposition tables I think
would save a lot of time for my algorithm
ye talk = executed on the spot 
Ah 
So... Execution is... What?
I think if my transposition table has a 16-bit hash index with 8 bytes for the full hash and 8 bytes for actual heuristic data, it should only take about 2-3MB in memory
that's not terrible
No really, What does that mean?
it's a de facto music channel right now, everyone is there to listen to music
if you were to go there and start talking over 100 people's music it probably wouldn't be welcomed warmly
ChatGPT is getting age gated 
I guess if it stops kids from cheating on school tests it could be a good thing but I don't want to give OpenAI my ID
Not that they're my main provider these days but still
they’ll just use gemini
Oh 
smart kids will just use the API
hi we're an AI company we love training data, btw please upload images of face-name pairings with rich metadata for us 
Smart kids probably don't need to use a language model to cheat
inb4 the api is age gated too
definitionally if they are using AI to cheat they probably arent all that smart 
you'll be surprised how being smart leads to laziness
good at programming ≠ smart
There's laziness and then there's shooting your intellectual growth in the foot
current child laziness comes with lethargy nowadays
people cleverly using AI to cheat are not smart
Yeah but they don't know that they're arrogant and smarty pants
A smart kid may flirt with the former but should know better than the latter
they are good at using AI and they are ensuring they will only ever be good at that
yup
it's a dentge move
is there a graph expert in chat? can that graph expert please explain how this graph is supposed to make sense
A lot of people in the cybersecurity space are quite smart, and ask any of them and they probably did something to somehow skip school, to them it's about breaking for personal gain
coin?
displayed numbers rounded to nearest 0.01
source values for line are not
/jk
hi I'm in the cybersecurity field, this is not true at all
thx
what do you do?
cybersecurity
not trying to dox myself
you're not gonna by saying what your angle is

small space sure but not that small
lol yes exactly






konii 2.0.1
how will i run firefox


using
5 miss



yea
