#programming
1 messages · Page 456 of 1
Size of my finger
"What happen when you move const"
These questions feel like getting punched in face
Then you're getting into a list of suspicious people
could you PLEASE elaborate on what you meant with that?
what in gods earth could you have meant with that abreviation???
competitive programming


Then intreviewer forcing you to solve problems, nobody expected



How do people even get swe faang lmao
Questions are insane
(Maybe just ask intreviewer for help)
I have nothing to do, so im random posting on discord
@safe path please fix channel
Please make me sleep
Shiro is admin tho
shiro?
shiro offline

alooo
I'm sorry im drunk bro
cloud please fix channel

what on earth is going on
start here
Major acronym issue
He talking about comptiteve programming
yeah let's not abbreviate that 😭
What did I miss 
explosion

Yes
this one should also go probably
Some chinese neigbhor have more programming exps than yall
what
insane esl


thanks cloud 
in a sense every person is a script kiddie because they had to copy from one another
wtf happened
a little tomfoolery
Acronym of competitive programming
HAAANKKKK
It’s funny until the second time you hear it
anyways who tf did it
the thai dude above
whoopsies
the thai dude doxxed his location, it was andrewbotics that did the oopsie with competitive programming
Dumb ruling IMHO, but OK
I guess autoslopping is not in vogue :D
in a competition about your skills as a programmer
you kinda dont want people to use things other than their own programming skills
what
we should make a geoguessr where neuro tries to guess where swarm members live /s /s /s /s /s /s
heavy emphasis on the /s
already been modded once dont want a warning today
Come to Indonesia and guess where out of the thousands of inhabited islands I am 
booking a flight right now /j
rtfyguhijkolp;'[/
(baaast)d
Mass Destruction
hi konii
hi t
hi konii
guys i have a question
do lsp servers process file entirely every time?
by protocol you send only changes, but how lsp work with it?
it is practical question tho
ye
ye
ye
z
bee zee zee zee zee
its gonna eat my dog
But the Estop is on the robot 
clear the area of all molecules, has to be perfect vacuum to function
any particles may apply forces that interfere with motion
hopefully there is a remote one on the controller too.
my robots all go into safe mode when i turn off the transmitter.
as long as power on - control signal always active
only issue if some one turn off control and go on conveyor
not fun to be that guy
- lags at least
||not long ago some one lost his arm||
i heard client can disable sending incremental change and just send whole file
if there will be no perf difference id just do this
language servers kinda have to do it at least once yeah? that's how they build source tree
then after it's depends on how lazy u are
some do, some dont
depends on lsp
like what will be diff for reasonably max 500 lines file
its common to have like 10-20k loc source files
maybe just disable incremental for layer and do pre-process on whole every time
i want lsp to answer immediately
between spending 2 days for layer and waiting additional 0.5 sec every time id choice first
discarded idea of changing rustc and rust-analyzer
what are you talking about 
it will be no any better than just external preprocessor and fake rust ls for client
when i change strung to string i want lsp to answer without thinking, or like so, for any change
traumatized by msvc lsp
reparsing the file is really fast
it takes like 1ms
parsing isnt the bottleneck in lsp design
why windows lsp for msvc from vscode think 1s then
That's Microsoft issue, not LSP in general
sure i use clangd and nvim native client
i know what speed is
yes 
clangd clangd clangd
slowge
missing sound
need mp3
-# stock
lsp bottlenecking 
for real, can i just disable incremental and no see no feel any difference?
c preprocessor without regex for just not not_eq or and tokens will be blazing fast
Ah my favorite
if rust analyzer will have to update internal tree for types and variable names, it prob will not be low cost
unless the file is absolutely gigantic id be highly concerned w/ my env if it was noticable
ok, i guess i discard half of what i wrote then :xdx:
so will it be better to complete internal file state and update line, not range semantic, or do whole file
and i think that's more about round trip latency rather than the actual parsing
whole file is simpler to implement because ugga dugga
the way rust-analyzer works is by just caching anything
even on reparse, it doesnt recompute anything if nothing actually changed
are you sure? i not sure, that why i, idk, concerning
incremental algorithms are possible to implement, but keep in mind that parsing is fast, it really takes like a millisecond
while incremental parsing is both a maintenance burden and requires more computational resources
one giant file
of course, i'll do it anyway
in my
editor
json repo btw
257M json/
dentleaf
NOT for performance reasons

280hz
freeeeeeeeee birrrrdd yeah
minimum
RIP minimum
ok fine
will parse entire file
now i need to look back and see what is actual structure on open/change/save
and is save even send text to lsp
and read about all other methods too
and when it will just work as layer, add preprocessor
caching is really the easiest method
then
uh
read salsa docs
i think thats the easiest entry point
maybe
iirc ra uses tree sitter which is incremental
though not entirely sure
i doubt it 
its possible
but i doubt it
i dont care if you use rust or not
read it
also
note that classic compilers are batch-mode, going through series of passes
while for lsp architecture, top-down (lazy, on-demand) evaluation is a better fit
i aint gonna create language or anything
i just need line-context parse for tokens 'not' 'not_eq' 'and' 'or'
(to use rust)

:owobred::ERT:
:owobred::ERT:
i hate this place
33 sold
didnt even bother to offer free shipping for one(1) stick priced at 3k
mfw 100k for 33 sticks of ram
this hook is user defined
i can just define it here
well i spend some time trying to find how to do this, and today chatbot just returned it
lmms are shit even as tool
give me back usable search engines 
charaverk | ~/.config/nvim/lsp ~$ cat clangd.lua
return{
......
on_init = function(client)
-- * magic chatbot generated type(), tho it works (for clangd at least)
if type(client.server_capabilities.textDocumentSync) == 'table' then
-- 1 = TextDocumentSyncKind.Full (Send the whole file every time)
-- fix lsp say that it is integer, but in runtime it is table
--vim.print(vim.inspect(client.server_capabilities.textDocumentSync));
client.server_capabilities.textDocumentSync.change = 1
else
-- If it's just a number or nil, force it to 1
client.server_capabilities.textDocumentSync = 1
end
client.notify = function(arg1, arg2, arg3)
-- ! and NOW i can catch didOpen.....
vim.print(vim.inspect(arg3));
--vim.print(vim.inspect(arg1))
end
end
}
i no longer need clangd_fake layer
i can just redefine notify
only need to code preprocessor
i guess in lua then
i mean
local original = client.notify;
client.notify = function(arg1, arg2, arg3)
arg3 = preprocess(arg3);
original(arg1,arg2,arg3, maybe arg4?)
end
fuck i wasted 10+ hours for nothing
i mean i got back to starting point
waaaaaaaaaaaaaaaa


Me at 00:41 am
is this true
yes
Also Evil
Thats more than my entire pc, which has 64 gb of ddr5
My killswitch is just the wall socket 
UK sockets have one built in
64 is more than enough for the vast majority of uses i'd imagine
that price is one(1) stick
I got 2 32 gb sticks and it was 1/3 that price. I have enough slots for 2 more if i wanted but my cpu cooler blocks at least 1 of the slots.
Streaming plex while using 50% of the rest of my CPU time + 100% of my GPU on training smolvla
Is this considered abuse
as it stands rn this thing will have 704gb of ram kekw
which will total out in this room to .......
128 + 96 + 64 + 32 + 32 + 32 + 704
1088gb of ram available across the machines in here

just hoping the terminators leave me alone when they take pity on me thinking im just the shittiest hyperscaler ever
One stick also has half the transfer rate of 2 sticks so it'll be slow
"half the transfer rate" is doing a lot of heavy lifting there. that's assuming dual channel consumer board, a workload that cares about bandwidth, and that the bottleneck is even memory throughput to begin with. single vs dual channel difference on something CPU or GPU bound is effectively a rounding error. if we're talking server platforms which is what those ecc sticks are then the channel count conversation gets a lot more complicated than "1 stick = slow." the xeon board in my closet has 12 channels so any combination in between 1-12 filled it'd be significantly more varied than that but yeah thanks
yeah so in that specific circumstance it'd matter more, definitely would hurt it but even then half of the theoretical max in a realworld use with all the other factors affecting it would likely not see anywhere near a 50% performance difference overall
https://www.techspot.com/article/3066-single-stick-vs-dual-channel-ram/ 12% across all these
apus that share system ram for the gpu in workloads that push that to the max would probably be the single most impactful usecase i could think of
charming art i like
it's very me.
I sort of try to put effort in, but I am just not talented with drawing.
The best art there is the Neuro sprite by Kotgedev
i'm not talking about the neuro sprite yeah
this unholy abomination of a game runs without any extra dependencies with python.
The GUI is TKinter :DDDD

Hey all, I've never programmed before and I started learning python back in February. I took Vedals advice and tried making my own AI Vtuber. It works good and is ready to stream but she's stupid, unfunny, and the latency is 8-10 seconds. If anyone wants to look at my code (there's not that much of it) and help me out, I'd super super appreciate it cuz I'm dumb dumb and have never done this before. Thxx!
That oughta do it
What are you using as the backend for it?
tkinter is a builtin module
And the hardware?
WHAT?!
in python
Huuuuuhhhhhh
MistralAI, specifically magistral-small-latest
Are you using it locally?
I'm on windows, recently got CUDA all working
What GPU?
That's news to me. An interesting one too
Frick no I have little baby laptop
geforce laptop rtx 4050
there are other ones
I don't suggest relying on cloud compute entirely
But as a suggestion, use the OpenAI API streaming mode to start off with, as it lets you get tokens as they're generated
Your latency however most likely comes from the use of a low priority low compute cloud API
Since you're already in the 30B size class though, using Qwen 3.5 27B could potentially fix the stupidness problem, I hear it's quite good
Yeah the model I'm using doesn't stream tokens so that's probably the cause for a lot of the wait. Dang ur right it probably is prety low priority
How much is OpenAI streaming mode?
Use Cerebras if you have $$$?
The namespace is kinda all over the place lol
You need to pass "stream": true into the API
Best speed
It should be a standard feature with any OpenAI compatible API which presumably what you're using is
Yeah I need to find one that can do that
What inference provider are you using?
idk I chose Mistral AI cuz it's so cheap
Do be aware that the demand for AIVTs is entirely saturated by Neuro so you can't expect to make a profit
How do I find that out 😭
Especially if your voice is like that of most other non-Neuro AIVTs
I chose TKinter partially for the challenge, partially since I do not want to package this, and this way common swarmite can just download python and press .bat file to launch it.
Just look at what website you got the API token from
Meanwhile literally every other programming language that bundles dependencies with the distributed code
That can run some decent models lol. I do everything from llms, to image video and looking into music generation on my rtx2060. Images take me about ~15s and 5s videos about 170s
Java JAR files are a great example of just putting everything into the release distributable
I don't want to write Java 
This would be far easier with built in java libs actually
No, no it can't
True
Kotlin better
Write Kotlin
6gigs are quite useful with the right models like said, i run a 2060, 6gb aswell.
Or swift now that it is available for android too
I run a 3090 and most models I can run still suck
A 30B moe works fairly decently on it for what it is. Its not SOTA ofc but yea
Like Qwen 3.5 is the one exception for models that don't entirely suck
Not useless for testing for sure
I don't want to write with mcdonalds language
it's been infected with mcdonalds since it's on their page
Tha's crazy
tbh, I could go for a chicken mcnugget right now
Do you think mcdonalds would sponser me 4 nuggets if I changed the project language to kotlin?
Well qwen3.5 is a thing tho, and even the older qwens arent terrible.. gemma also is decent, espescially if you dont need perfect accuracy. Id say enough for a streamer start, and you can offload the big important stuff to a API anyway, and have the small model be the latency "bridge"/filler
This handy little thing runs on Kotlin
Anyway, all i was trying to say is that a 6gb card isnt completly useless or incapable of running anything fun
You just compromise quite a bit
I am dumb and I always make this kind of logger myself for my projects,
I just like writing the logger part for some reason idk.

(and yee, streaming tokens is likely better for this)
Definitely won't be training any vocal synthesizers any time soon tho
Or train anything more than basic ML, yeah.
Yeaaaaa
6GB is not amazing
As a user tho, its surprising what u can squeeze out of it
4gb gets tight, and often slow..
How to sleep
Qwen 3.5 27B rocks imo
Gemma 4 31B will rock if they do a 4bit QAT release like they did with Gemma 3
Right now it's a tad impractical
It's like one of very few that don't suck
This is true
I would say it's the first model in that size range to really feel like it comes close to the budget cloud offerings like Gemini Flash and the smaller ChatGPTs, as well as some of the older frontier models
ctrl alt delete then hit lock
the game just keeps expanding itself with the extra windows.
soon it will eat the whole screen
(each of those little boxes is closeable, so you end up with just tiny corner game when you don't actually actively play
Yeah except I'm built different so
Surely
I hope you have a good quality voice
The voice is a big part of the experience
One big issue is that even if you end up making something more fun compared to Neuro
it doesn't really even matter
The audience wont move
it has to be genuinely far far above Neuro to start moving audience there.
Do you think it's better if it sounds super believable, like a human, or more robotic like Miku or Neuro?
depends on the character personality and how the vtuber model looks
bruh thats up to you man, everyone's gonna say different things
voice is important part of conveying character in this kind of space, everything has to match
As long as it's not unpleasant to listen to
Avoid sharp high frequencies
It'd have to cater to a completely different audiance entirely
probably
I messed around with qwen tts for a few weeks, it can design voices super super well. Super customizable. Can even combine ones.
Only one downfall
It cannot do non-enligsh native accents accurately
you literally said "how do I find that out" as a response to "what's your inference provider", you are built different, and that would make it at least very hard for you to do this
Lmao cuz I've never done programing before in my life, started in February.
Literally no idea how anything works
you definitely should learn before even trying then
well idt you're gonna get much help here, at least probably not beyond some basic stuff anyways
insert #programming faq here or smth idk
I was just trynna get someone to read my code
yea I thought so too
also i don't get what the API segment part is about
@amber fractal dinkdonk
^ Just asking around for a better LLM and wanting suggestions for latency.
Iggly how dare you still exist in this channel.
You didn't support me during the Harpoon Wars of 2026, therefore you are my eternal enemy.
lmao what
you're orange tho. you don't deserve support

best I can give you is try running locally if possible and cut down on unnecessary wait times between input and generation/tts
If he gave me some more harpoons during middle stages of the war I would have won
harpoon spent to kill an innocent is harpoon well spent.
-Evil (probably), 2026


that's my honest opinion
on...?
understandable
how do you feel about the html in canvas api
idk what canvas api you talking about but html in api doesn't sound like api to me
it dropped recently as an experimental flag in chromium canary, it basically allows you to draw call html that's already on the page
here example
can't wait for non euclidean text selection
lol
one is just sparkles, the other is a sparkly pencil

schema store issue?
nvm it seems we can't use tsconfig.app.json now https://github.com/SchemaStore/schemastore/commit/979bb17b767d05ebfb26a22f903cba117c7c339a
- Schema URL: https://app-protocol.org/schema/app-1.0.json
- File pattern: *.app.json
- Description: Structured JSON format for professional profiles, resumes, and CVs
The Applicant Profile Protoc...
I don't know wtf this is about but 
Felt lazy initially when I created these sub windows; they were spawned directly with functions in the main game loop that is supposed to handle the main window and the content happening in it.
Feels good to have refactored stuff now that each window has their own class based Core class that they inherit.
I can actually read the main gameloop again.
There was hundreds of lines of code to initialize different ui elements ect for each sub window there in the main game loop, and if I continued with that technical debt I would end up with idiotic mess that'd take days to clean later so I dare to publish the code
TLDR
Finishing refactoring code that you know will cause you issues later feels nice.
Yes it does
I'm having trouble thinking of test plans for a project I'm supposed to be making a proposal for
like at what level?
do they want the whole testing pipeline or?
just jenkins it up and speak to them about how good AI is at writing robot framework tests (it's not, AI is horrible for complex test writing)
its my bachelor's capstone essentially
I'm writing an implementation of ML-KEM and ML-DSA for IoT devices, and I can't think of what scenarios to test and what outputs I should look for.
Try testing at every level - even if just to verify functionality. Unit, integration, end to end, heck, throw in some load tests too
@amber fractal i have a nit for the voice section specifically, “voice cloning” tts models are commonplace now; the qwen one is popular but there are multiple others that are pretty solid - my whole point here being to just let go of rvc, it’s ancient at this point and requires a second inference step, and isn’t even necessary with these new models. “llms doing vocal out” either refers to that - tts on llm-based/adjacent architectures e.g. gpt-sovits, llasa or basically anything that comes out nowadays (these just use the llm part as a fancy translation model from regular text to speech tokens) - or omnimodal models which are either proprietary (black box, can’t tweak) or dead ecosystem wise (looking at you, qwen omni series when you were still open weight)
as for stt models, there are newer ones which all might be a very viable alternative to whisper, notably nvidia’s parakeet models
+bonus nit 
for vision, object detection is not a viable solution as it will not do what is expected from an llm that can “see”. for the purposes of making neuro a vlm is basically required. probably even one that can take video
Usually border accepted values and couple values from middle are safe enough.
Longer test suites that test everything are pretty standard these days.
But they are only run nightly instead of whenever merges are done.
So units and simpler tests should be done for all commits that are going to end up in master.
complex tests that actually use the product can be ran nightly ar at demand level.
load testing is usually done for actual releases. During the release period.
If you're planning on doing just single release then having it be part of nightly tests during development is fine.
well as you're going to be writing stuff focused on security.
The main issue is displaying the feature performs correctly under expected use of the device, and when the device is under attack.
Tests that display trying to fetch data from the device that would reveal vulnerability is important, like malformed messages can't somehow bypass security requirements etc.
Forming connection to the device by malforming the message. MQTT and WebSocket connections can sometimes allow this.
got jumpscared by a reminder from cloudflare registrar
holy yap

Mostly malformed packages should lead to crashes instead of security issues, but security features should in my opinion be able to handle attack situation like this smoothly if possible.
So testing different kinds of malformed packages can be good.
from development perspective aswell
your tangent on attacks is interesting, but I don't think that malformed packets are much concern for a cryptographic algorithm - if the data's garbage it's likely to just reject the packet or crash I think
however, vulnerability to a DDoS attack can be important as well, but I don't know if the code sandwich that I'm supposed to be envisioning the implementation it in would necessarily have the ML-KEM or ML-DSA algorithms be the relevant function to handle that problem
if you only handle the cryptographic part, instead of whole authentication handling from when the connections are formed, then it is likely outside your scope of testing
hm, DDoS attacks on something using ML-DSA may be something testable
and I can't believe I invented a term that's the opposite of spaghetti code
unboiled dried-spaghetti code?
code sandwich
code that has an obvious causal relationship between functions, simple, readable
but what if I sandwich yakisoba?
or in this case the whole thing with how encryption is layered into the process of network communication
disgusting
How is it looking like actually for classic encryption methods RSA and stuff?
like one reason to do what you're doing is that people think traditional methods might be / will be become too vulnerable.
what is the state currently?
cloudflare claims that RSA and ECC are vulnerable to specific quantum algorithms, but no computer exists (yet) that could exploit the vulnerability
additionally I framed the project as an experiment done because in case one encryption algorithm is insufficient there's another that would be available to replace it, doing the work ahead of the exploit rather than waiting for it to happen
Big corporations are very interested in these kind of things sometimes
The amount of money it costs to do this (fixing security) when it becomes actual issue
is astronomical
so you might have actual good prospects with this research if you do stuff well. Corporate loves not having to spend money, and are willing to pay money to save a lot of money.
if anything this makes me understand why bank atm machines are so fucking slow
Isn't there security in slow responses themselves.
Hard to try to break anything when you get one response every 5 seconds.
Hard to break something already half broken

ptxas error : Entry function '_ZN5bloom6detail20insertSequenceKernelINS_6ConfigILt31ELt21ELt27ELm6ELm256ELm768EmEEEEvPKmPKjS5_mS5_mPNS_6FilterIT_E5ShardE' uses too much shared data (0x13800 bytes, 0xc000 max) the bane of my existence 



i guess i can just increase the % that's made available as shared memory, surely the lack of L1 space is going to be fine 
gguys i got a question
the neuro's osu ai, was it trained with only vision or did vedal use the map's data
im not sure
Idk if that was talked about in reddit but im too lazy to check
that looks bad
it means amd memory controller hates me
unless a stick died somehow
i dont really feel like troubleshooting this
reduce memory speed, increase SoC voltage
if a stick didn't die then that should probably fix the issue
SamVanTriangle: infinite cheap GPUs, infinite free RAM problems
While i was at it i undervolted my cpu
That make sense. Thanks
Free ram? In 2026?
RAM issues
Ram always have issues.
Don't panic yet, it could very well be that you're just using the wrong expo profile
Im on xmp1
1st stick i tried failed.
Lets see the 2nd stick
2nd stick passed 
Surely changing to a diffrent gpu shouldnt affect ram
If it has different behavior on power ripple somehow mabe?
Very far fetched tho
The 7900xtx uses only 5W more tho 
3rd stick passed
4th stick fails
2 sticks failing would make it either really bad muck, but its 99% sure the mem controller
Welp, upped the voltage from 1.40 to 1.45
Still fucked
Fuck it, no more T1, it will probably work with T2
okay before you follow anyone elses advices

what you did beforehand?
In terms of debugging?
Nope
Yep
Ignore this, i completely disabled PBO and everything just to make sure that wasnt the issue
2dpc?
4 sticks ye
amd's am4 imc shouldn't be bad
2 of them failed even when only running 1 at a time
why is yellow name so popular in #programming is it like faction flag?
oh
banana.
Banana campaign
but still boots somehow?
Yep
okay i'd rule out playing around signal amp
Maybe i cooked the mem controller by setting soc to 1.17v 
i noticed some quirks
if you noticed, on all the value comparison
the 8th value from left side is the one always being different
just in case, when you individually tested the ram
which slot do you put them in
can you maybe test both the failing ram
and send screenshot like this
Sure
its really odd when the failure is always on that one specific hex value
e - c
8 - a
6 - 4
9 - b
a - 8
3 - 1
7 - 5
0 - 2
okay its hella suspicious the difference is always 2
the ram itself possibly
probably knocked off the emi filtering cap
or
according to what i see that its consistently off by 2 in the hex, completely zeroed or one trace on the ram
but i need both to be tested
its the 33rd byte possibly
no one in here seriously noticed this?
really tells me this isn't signal integrity nor addressing problems so rules out xmp/expo/voltage/timings stability
I wouldnt know that is out of the ordinary 
even if only the 33rd byte has integrity problems it should've been random
but in this case its always fail
this is the first stick
I dont think i have a missing cap on my mobo
so im assuming its 50th and 51th bit
odd that it just never offsets 6
Maybe interference between traces?
so the 2 bits flipped randomly taking turns
okay
have you tested other mem testing software or like tested running something on them
the problem is so bizarre i seriously doubt its hardware fault
unless if god for some reason decided it would've been so funny to randomly add some funny xor logics into your ram traces just to fuck with you
I didnt test other mem test software, but duscord does crash a lot in the last day, same with firefox.
On the bad sticks i also judt had hyprland not booting easily
Sometimes had to go to tty3 and switch back to tty2
2nd stick
took me a bit cuz i didnt remember which of the sticks gave issues
so its consistently like that
😭
okay
now im suspecting its the internal dram controller issues
where it accidentally offseted something like the trace or some shit
and specific to that exact bit i mentioned
you said something about the rams ticks maybe misisng a filter, but im not noticing anything missing comapred tot he workign sticks
oh
the ram's internal controller
FAILURE: possible bad address line at offset 0x0000000089cedc8
When an address line is bad, the CPU tries to write to one "mailbox" in the RAM, but the data ends up in a different one (or nowhere at all) because the "wiring" (the address path) is faulty.
less of address and more of possibly during fetching it messed up specific bit
possibly wiring gets fucked
The RMA "Dud" (Most Likely): You likely received a "Friday afternoon" kit 
why does ram keep breaking for me?
this is so unfair
you're not the only one
except in our case it might be just that 32gb dimms are so ass to work with
just to make sure. the ram was still int he packaging and everything so i had assumed it would be fine. but the post office did put the ram kit through my mailbox thing in the door and it probably fell 90cm
woudl that cause issues like this?
they're all 16gb
i wonder if is till have a screenshot form my old issues
if its the same that would be fucked
but well that's the furthest i can help with when everyone elses "try different xmp/voltage" advice doesn't work
oh now i'd really like to know
what are the chances they just sent back the same kit
or they repaired it but messed up some bga shit
im pretty sure this aint even addressing issues
so its the data lines
i thought they sent new kits instead of repairing
didnt find screenshots
but found pictures of my screen
Seems to be the very last bit
this was on only 1 stick
and it failed on any timings, even very loose ones
lemme check if this one works at lower timings
So specific bits being wrong is a reocurring issue
I wonder why some of my ram has it and some dont
I thought it's called a "Monday kit" lol
friday afternoon is whn you dont care cuz its almost weekend and you want to get of work
monday is when you're grumpy
In Germany we have the sayword "Montagsmodell" which is basically Monday model / Monday kit 
The API is a wildcard mostly, could be split into games and other programs
It wasn't a lust for old tech, more like RVC was all I really knew existed well, RVC might still make sense in a more custom area but it's fate is always the same; getting merged into the LLM
Learning to make an rts-style point and click character movement in godot
I'm crossposting to here
man, this is painful to do.. why did I make it under 1MB 
to have remotely listenable audio, that only leaves about 250kb for the video 
which, imo, is about 24kbps, no parametric stereo
the good thing is the encoder behaved fairly well here without having an aggressive sbr start (which can add a lot of noise and remove important harmonics) and also 11.5khz of frequencies (I'd like 12khz... but oh well.. can't change the lowpass with this encoder)
New color drops 

perhaps just some gaps in the frequencies 
why did you stick to the cooler?
If it was in there that tight then the noctua cooler being the chonky boy it is, would have been pulling on the cpu
true
Hey @fast pagoda
Tbf I need to work on the other branch with the best name I've come up with in awhile
New colour roles
Thankfully there's a github with dota-style pathing code
Done that to a PGA CPU before. Helps to warm it up and twist it.
Should be fine
where pins
No bent pins

Went nicely back into the socket
wait, I think I screenshotted the wrong one
My hands and besheets have thermal paste on them now tho
just dont get it in your eyes
you will get eye cancer instantly
it definitely had a bit extra coming out the sides lol
the eye thing isay because i was just cleaning some paste up and read the warning on the paste i had
and yeah it is basically like
ur eyes will fall off bro
do NOT
get it in ya eyes
I will try
yeah, I did
went ahead and lowered the resolution to make it less blurry, too
you just need to change to mp4
Stream #0:0: Video: av1 (libdav1d) (Main), gray(pc, unknown/bt709/bt709, progressive), 192x108 [SAR 1:1 DAR 16:9], 1 fps, 1 tbr, 1k tbn
Stream #0:1(eng): Audio: speex ([9][161][0][0] / 0xA109), 8000 Hz, 1 channels, flt, 2 kb/s, start 0.007000 (default)
wtf is speex
mp4 doesn't support the specific audio codec I used(speex)
kinda similar to hybrid encoder opus, I think?
but more listenable, lol
just tried hard-cbr 2kbps opus, it's bad
❯ 7z a -mx9 output.7z output.mkv
Creating archive: output.7z
Add new data to archive: 1 file, 194775 bytes (191 KiB)
Archive size: 97133 bytes (95 KiB)
We can go smaller
smh archive file doesn't count
I like this audio better
but it may be slightly higher bitrate, idk
where are the middle bands
the encoder is whistling Listening
sounds like someone whistling along in the shower 
that may be why it sounds that way 
a lot of encoders don't support 8khz sample rate
but raising that is going to eat a lot of bitrate :/
Vorbis supports 4khz sample rate? waaaat
that's a shame, this is as low of a bitrate as I think it allows
I want something compatible with Discord 
blue

Blue = ReactJS = Hate 
aintneurway
i may like the blue
and the amethyst
Wrong blue
React blue is similar to aqua name colour here
like this
yyou can give me the colors too
Oh so we have aqua
That was always there
ask hatsune miku
I always had that on because it was the closest to blue I could have
I know what she likes

I didn't know lol
bro was hating on blue
konii
mmmmmm...
the blue is a nice shade
this feels like basically gray
i didnt like it much at first

i not
damn i might have that same exact heatsink
I think @hushed creek is lost everyone, you think we should help?
neuroMlem
until i think of a new serverwide campaign
nah dude just need some 💊
probably
her body so warm
Touch grass campaign. Be green
scratch that, definitely
Hi konii
since we have more colours it could be something like "open ocean"
Open forest
I'll be part of the Great Barrier Reef supporters with this blue
Konii, why does ram hate me?
Does the triangle make me deserve ram death?

although idk if aqua blue would fit better
surprised the gpus haven't bestowed gpu death upon you
cap, show your miku merch
nice embed

But I do make sure she doesn't talk to anyone else

Oh this is the programming channel
sir, the schizo channel is right over here
ye channel for NOT baking user
no this is the technical channel
guys why am I glueless
guess who heavily uses macros in C
except for me
Says the biggest nerd
everyone who wants to pretend polymorphism is real and not void* ptr

I love failing ci
what did OOP do to my guy
i fucking HATE oop i fucking dont want to use it
??? bro it's been preserve module and bundler module resolution since the beginning wtf?
why, may I ask?
i dont know i just dont understand them even if how hard i try
dang you must've be using Python
oh...
what do you not understand about them?
the entire thing?
it's just parents having a bunch of kids
i understand it --> next day i forget it completely
yeah something like that
its more of a personal skill issue
but this occurs more to classes
never occured in cs to me to anything else
what are vtables, sounds pretty good to have a hook on 
vtables are related to oop and are a form of rtti
rtti means
runtime type information
should I stop signing my git commits with gpg and just use ssh for it
I feel like the "standard" gpg client for windows is too slow atp
but at the same time I don't want to have to unlink my keys from my account because otherwise all my commits show as unverified
what does it have to do with code mutation
new Human()
am I 
self-modifying code
no
once you have ability to read and write arbitrary values to memory addresses you can modify a program at runtime

concessions of allowing more freedom is also allowing more badness
only ram has it out for me it seems
but Awa doesn't have a memory space
weirdge
only in von neumann

Considering what ive seen happen on my laptop, i treat my desktop gpus pretty well
That 3050m has seen horrors like no others
I mean I can make a dereferencing feature, which dereferences a register value, and it'll point to a value on the stack
ld
also
wrrrrr
Awesome self-modifying code vs evil and intimidating page table
so, an instruction that moves the Xth place value on the stack to a specific register

@true hemlock
im losing braincells each second
so i went from 2 broken sticks to 1 broken stick
is it a software issue?
surely its not a software issue
welp, imma retest all sticks
But it's clean 
Isn't that functional the programmer's final dream. A function that only transforms data
and it fucking works?
3 out of 4 now
what's the result on the broken one
Still same looking 1 bit wrong
no?
erf
Yes, because I say so 
Do normal functions not change data?
Sam, killer of ram 
functions can loosely be classed into ones with side effects and ones without side effects

ones without side effects are better behaved

New idea
The soecific data lines that fail have slightly higher resistance or something
And the 5950x hates it
Im testing on the 13400
wrrrr

at this point I am trained to assume the box is the emote from my name

Its not
Even my phone cant see it
13400 works
Lets see if the ram works
The stock ram is micron 

The ram is def still broken in the 13400
how does THIS happen
Wdym how? Of course by hidden quantum fluctuation energy!
In all seriousness, it is the charger detection that is wrong
the phone charged on credit knowing it'll be able to recoup the losses soon
The charger is plugged and giving power but the sensor didn't detect it
how did it charge linearly in a perfect line
while not plugged
also its a laptop not a phone
do they do smoothings
its probably cause of that
it was plugged in but fully off because the battery died
they always seem to do linear increments
It's also possible the max charge is capped so it does become linear
so it was missing data and the smoothing algorithm prolly just made a line because of it
9 | 9
uhhhh what
there's a national holiday called "a" 
Oh what
They typod lmao

I wonder if its an automated message or just one guy that didnt get a break
24V rn, 2.4A













should I implement pointers for Awalang



i put so much effort into that essay