#programming
1 messages · Page 383 of 1
Because if 60cents/minute is with assistance, then that is cheap af
i doubt it tbh
they jsut have the steps for how to use it ont heir website
nothing mentions assistance
Well, at least 2D cnc gcode is not that hard to understand
so worst case a few dollars while you work out what to do and do a practise cut with a similar material and then do the actual thing?
Depending on how big the cut is, it may take more than a few dollars
ye
i forgot how enormous a 3090 BAR is
jsu tmake the bar resizabel
chat
are we unc enough in here that we've watched stand alone complex back then?
resizabel autocorrects to Elizabeth 
did it motivate us in career choice?
but the msi's trx40 suck ass
cant even correctly assign a PCIe slot mode
when rebar
had to do it manually
- rebar and 4g encoding
like the anime? nope
also no
You should know the machine recommended feed rate and power for the thickness of the acrylic that you're going to cut. They usually have a table of it. From that feedrate and gcode, the gcode generator should be able to tell you how long it will take to machine it
Usually acrylic is 10W/mm of thickness
Yeah, but the feedrate will depends on power too, and that usually depends on the machine (like how close is the toolhead)
Feedrate is just how fast the laser should move (in mm/s usually)
well you see
i have no clue
you're getting ahead of yourself here
or ahead of myself i guess
The topic of machining is exciting to me because I did that for a living lol
yeah what the hell
nowhere, it's just memory-mapping
would your design let you use the untrimmed acrylic piece? because you probably want to keep your options open for how close to the exact size you want to cut it.
i can cut a 2nd circle out of the other side of the lens in theory
its literally just MMIO
but still
i need a fresnel lens, but the optical center has to be at the edge of the circle
16GB is crazy
still seems tiny tbh
not even enough to map all VRAM at once 
PCIe restriction mabe?
Wait, you're using acrylic as the lens?
elvyns threadripper been having PCIe issues lately after she has the 3090. turning on reBAR, 4G encoding AND manually configuring the PCIe slots fixes it
the fresnel lens ye
im not able to cut glass nor quartz
BAR size usually isn't even that much
most i've seen is 256MB
the 3090 randomly getting negotiated to have 16GB addressed is insane
It involves a lot of water
Or if the glass is thin enough, then there is a specialized cutter for it
glass cutter probably wouldn't work well on a fresnel lens because of how they're shaped on the non-flat side
Yeah, that's the thing. I don't know how optical glass is shaped
whats wrong with acrylic tho?
it's got a lower refractive index than glass
Nothing lol
i think
i think it shoudl be fine
ye, because usually you don't need that much, it just slightly accelerates transfers (and would probably make development much easier because you don't have to remap stuff all the time)
though I'm pretty sure DMA also bypasses this
I don't know enough about optics but you're saying you're unable to cut glass. So I'm just telling you how to cut glass
i jsut dont want to deal with the hastle of cutting glass
i still was expecting at most 4096MB of BAR
and got 16384
and acrylic was a lot cheaper
i think large is good no?
honestly
K80 might also default to 16GB BAR
right
last time i tried putting in the K80 on my TRX40 same issue with debug hex code looping happened

i wonder if the board will survive the 3090 + K80 together
I think it just takes whatever it can get (because as Sam said, there's really no downside to using a bit more virtual memory as long as there are no conflicts)
also apparently my GPU currently has a 32GB BAR 

what
kind of gpu is that
7900 XTX
codex didnt like this prompt "avoid search tools, line by line, everything, update doc after every file. src/audio src/debug src/ui src/expression src/ filters src/gates src/gpc src/llm src/memory src/orchestrator src/session src/streaming src/stt src/tts src/utils src/discord_main.py src/main.py src/server.py"
imagine vibe coding lol
you can have a BAR of up to 128 TB
upper limit supported by the Resizable BAR capability
Id rather go to sleep then go through thousands of lines of my own python code
python?
damn
i would really like to say low BAR but shuni appeared with the info about BAR upper limit being 128TB
right, seem to line up with 48 bit addressing
or 47
though hmm
they just ran out of bits in the configuration register 
this isn't the address itself, it just gets read to know what the device wants/supports
i haven't actually looked into this
ye I haven't either, so I was curious on how this negotiation works
smh just use those 4 bits
apparently it's just this register
and the host picks whatever it wants from that I guess
wade
they added another register 
so yeah, 8 EB BAR 
That's rare
tfw pcie 5 spec from 2019
It's like finding random ISO spec in the wild lol. I think either someone is in trouble or they just don't care
They are paywalled by (relatively) a lot of money for individuals after all
They want so much money it's ridiculous 
$4500 pdf 
Yeah, welcome to standardization. Some spec are ridiculous for what they are lol
"standard" as well is crazy
It is what it is lol. Automotive industries are very notorious for it
Hem
I wonder if mine has any
How would one even know how much memory a GPU has decided to map
i getting async work when it impossible to get
unless c api use some sort of suspending
so, you want to tell me, that inside of c function, they suspend execution on some point, and than return to this point, while allowing any other code be executed on this context??
what the fuck
That just sounds like OS scheduling
coroutines 
cpp main
-> cpp qthread
-> execution context on qthread
-> c api function
<- somehow return execution context ownership, idk how to call here
-> execution of next function on this thread, from queue
(blowing everything up because it should be in defined order)
->c api function get control, do their work
<- just exit from function it was called from
<- and high lvl context know nothing about it
that the fucking reason i so pissed off
there no any
it fucking c functions
There's no coroutines in c (by default)
i gonna safely leak some code
wait a moment, i need to download browser and login discord
is it one of the nonblocking options that were set
there nothing in docs about this behavior
void Api::requestChangeActiveMode(bool active){
qDebug() << "how tf you run here";
qDebug() << QThread::currentThreadId();
if(not isWorking){m_active = false; return;}
if(manager->isValidCamera()){
manager->activeMode(/*bool &*/active);
}
m_active = active;
}
void Api::requestResetClient(){
qDebug() << "i run here";
qDebug() << QThread::currentThreadId();
// std::this_thread::sleep_for(std::chrono::seconds(10)); // checked that it is indeed preemptive c function
try{
dropRuntimeSizes();
manager->closeCameras();
manager->startupCameras();
if(not manager->isValidCamera()){return;}
manager->setSettings();
if(not manager->isValidCamera()){return;}
auto pair = manager->getSizes(); // ? size of channels variable, not channels of camera
changeRuntimeSizes(pair.first,pair.second);
std::string text = manager->getSettings();
if(not manager->isValidCamera()){
dropRuntimeSizes();
return;
}
changeTextOutputInSettings(text);
qDebug() << "camera ok";
isWorking = true;
}catch(...){
isWorking = false;
std::cerr << "error in " << std::source_location::current().function_name();
}
}
i run here
0x6170
**request active mode <- outer code**
**startup cameras** <-inner cout**
how tf you run here
0x6170
UsrMsg Camera error in initialize : "Abort" // aborted by myself, but in general it just continue execution of requestResetClient
btw, all names are legacy api
But yes, it does require libc
unrelated generally
it not even linux
i have to use fucking windows for 3/4 apis we use
ucontext from the man page is from standard c library
aint no way it can non-blocking non-wait-point non-external hidden-from-user call async multithread scheduling
but seems like it is
That's what it said. But I don't know if that truly is how it is because I haven't tried it myself lol
it behaves just like it use some sort of poll, but without waiting
if it is thread that wait on interrupt...
nah there no use, just interrupt cannot switch context back
how tf it async in first place, how tf it can behave like co_await
This smells like qt abstraction magic to me
qt just do qthread{qobjects_linked,queue}
and qobject.forThreadAdd(task)
ot something like this idk
but here is JUST function
outer code know absolutely nothing
it is compiled dll
I've never actually used qt so idk, was just a feeling 
no, qt here is just like threads with async scheduler, who can assign task to specific thread

guys
i give up
they claim that they use qt4/5 core libs and do not interact with user code
qt cannot do coroutine
qt can invoke process incoming events (slots are events)
but
how tf their dll code with own, assumed statically linked qtcore
with FUCKING c api functions
can interact with outer (user) code
that is only reasonable explanation
but that fucked up
totally violate some standard requirement
idk which one, but feels like it do
/
msvc is Worst compiler
maybe it is a reason
gonna add timeout 10 sec before calling request active mode from outer code
just some sort of execution guard looks better
and also JUST HOPE that no one will ever try to add something that should behave sequentially
^ should be in our docs at least

well, this sounds a lot like some sort of undefined behaviour or corruption
if you want to dedicate more time into this, make a separate minimalistic app, just enough code to investigate
it sound like
but it is not (my part of code) ||// at least||
btw, and even worse, even with queued connection (macros always expand to queued), it still dead lock itself (as you can see here
)
i have to add qtimer::singleShot(10,... here
anyway, after understanding issue it took only 2 hours to rewrite related code and end testing
now, i again have no work to do, until coworker finish his part, and than there is like at least 2 week non stop brain-paired work
time to watch youtube
Well, if that is a highshool homework then the teacher will probably ask how did you come up with those numbers to check if you understand how it comes to it
But knowing ID school, probably not
pretty sure they don't allow desmos for most tests either.
Homework is not test tho. So it may be allowed to use calculator but you still need to explain how it comes to those result
But if that is uni, then probably the prof. WILL want you to write the explanation
explanation to show that you didn't just buy the teacher's edition of the textbook
i think
CS teachers should ask
where they found sources, why they think that function behave the way they expect, and etc.
I don't know about other uni, but at least my uni is not like that. In fact, the lecturer on my program strictly said that you are better off downloading a pdf from some site
Yes, exactly. If it is uni, then the prof/lecturer will want an explanation. But highschool? Not so much
obviously? 
its a linear regression problem, solving it with desmos is like looking up the answer
Or it could be the usual mindset of "too tedious to do manually that may take hours, spend days of automating it"
"I didn't get ChatGPT to solve it. I asked ChatGPT how to use Desmos to solve it."
Honestly, I assume the person knows how to use desmos already
the only math class i ever took where they wanted you to use a tool to do it was one where they taught Mathematica.
My class in uni allows us to use calculator all the time. Geogebra, desmos, scientific calc. You name it. If it helps your understanding, the lecturer allows it. At the end of the day, the test will not allow you to use all of those, but maybe some table if it is statistics
Yes, even for homework
Or calculator. Calculator is still allowed at test because I vaguely remember borrowing one because I forgot mine
was only calculators. no network stuff.
if they have to show work, i think they are not just cooked they might have boiled dry.
That's what I'm saying. They have to show work but if they understand where those numbers come from, then it is the form of "too tedious to do manually, automate it instead". Because I assume the person knows desmos, I think they at least know basic math
I didn't know desmos could be used to do linear interpolation for example
So that person may have been using desmos for a while or search it up online. I assume the former
i've only ever used it to visualize functions, never for analysis.
but I know it's powerful coz someone made Bad Apple in it.
tbf, my teacher specifically instructed everyone in my class to use excel (so that we wouldn't have to calculate the regression manually) but fuck it I decided to use desmos instead lmao
yup, I know how to utilize desmos for the types of questions that need it
Then naah, you're good. Just put the explanation of how it works under the hood. If your teacher is specifically teaching how to use excel, then you're cooked
nah, my teacher be like: "class, just use excel for today's hw so that y'all wouldnt have to go through the pain of calculating everything manually" without actually teaching how to use excel for this type of math problem lmao
thats why I used desmos, cuz im already familiar w/ it.
Yeah, I assume it is uni stuff? If the lecturer is young they're usually more chill about it
nah, senior year hs
Even better. HS teacher usually doesn't get paid enough to care in ID 
and most of em aren't aware that desmos exists lmao
im totally gatekeeping ts
Yeah, quite sad really. It is a good teaching tools. Not much video got translated to ID either from creator like 3blue1brown, or numberphile
Language barrier sucks lol
if they knew desmos exists, it would get banned asap. cuz they'd think desmos just freely giving out correct answers to students. Even tho thats not true
Naah, that depends on the teacher. But knowing our country, the majority of it will do exactly that lol
My hs teacher used to ask me how to make interactive lesson. At that time, I only know manim and not desmos so the barrier for her is quite high
But she did use geogebra
if only i knew desmos as early as starting hs, I'd be the straight A student cuz of desmos lmaooo
no shit, even UTBK's (Indonesia's national college entrance exam) questions are solvable using desmos lmaooo (not all of em, but the majority of them are)
I came from a time where national exam is still a thing. And I choose physics as my subject, so that wouldn't help much 
It was either physics, chemistry or biology. I suck at memorization so there is only one option
obviously, cuz that's outside of desmos' usecase
Although, lowkey, I think SAT questions are a lot easier than UTBK's questions (even tho both are solvable w/ desmos)
out of all my past classmates barely anyone passed UTBK
though i don't really know myself how was it because i got to skip 12th grade and went straight for international uni and left indonesia
but oh well indonesian education is fucked regardless lmao
i not just cheated by asking LLM to find it, i just used NSE instead of WSE
how long time ago you left Indonesia?
last year, had to do college first in there but it was still owned by an international uni
bru, I barely managed to get a 500/1000 on my utbk prep tests. Meanwhile, I'm getting a consistent 1300+ on my bluebook practices (for my SAT prep)
SAT felt way easier yeah
R&W lowkey a pain tho

is country growing or declining?
i not know what is better place to migrate to, i also look toward low-but-growing economy countries, i gonna (search for) remote work anyway
economy sort of worsen though i'd not willing to go through the politics
though almost everything there was dirt cheap
NZ and Australia will be countries that i want to look for, but in first place, you have to find a job there
depending on how good you are at finding deals, $3 could get you 4 pieces of fried chicken and 3 rice balls
but income there also suck
Australia's the opposite. high income but also high living cost
i gonna (search for) remote work anyway
rn i just wanna leave because of censorship and sanctions
also its pretty hard to get into Australia for job
visa's been getting rejected even on many institutions
the opposite. high income but also high living cost
that like, everywhere
workforce are barely accepted
except for third world countries which are... quite a lot
that is why i consider different places
If you managed to get remote jobs from high income country, ID is not a bad place to live
Because for high income people, the economy doesn't affect them much
average people in Australia made a month worth of average income in Indonesia in just a single god damn day, sometimes half
i mean, if high economy, than median income and costs reasonably high as well
and opposite for mid and low economy, except countries with broken economy, not many of them now... will be more later on
Put it this way. I know a guy who is a welder in czech. He lives very frugally there but can make enough to send to his wife and 2 child until they graduate uni
In Indonesia
yeah the income difference
is indonesia economy growing or stagnate?
its insane how i could just work for a day or two here in Australia and paid like probably a semester worth of my little sister's school fees back in indonesia
Declining 
declining
absolutely declining
its only a matter of time before indo fullly collapses
Hence why I said if you managed to get a remote job, it is not a bad place to live lmao. For high income people, they're not affected by much. Maybe spend less on luxury like brand fashion and stuff
i just push research further more from now
rn i have to make visa card and get a job in $ or euro
Indeed
if you could get an international remote job and live in indonesia your life would be so damn easy with how cheap the living cost there, given that you don't spend your income lavishly
who need luxury, i want just place to sleep and shower, free and direct access to information, food,, and at least reasonable censorship (btw australia is questionable here)
lmao
That's what I mean lol
yup, indo's by far the cheapest country to live in
well i'd say indonesian censorship isn't as bad as Australia rn but they block literally anything they thought is a matured site
including reddit for some reason
but not twitter lmfao
Indonesian government doesn't censor much. Like, they are a facade so to speak but rarely enforced. If you speak against them however, sometimes they are enforced if it got enough attention
i can open reddit tho
i just visited my home town not long ago and i actually felt rich for the first time wth
They block it, but not hard block it. It's as easy as changing dns
yeah
Or even using dns over https
but indo's minimum wages suck
what would you expect on the indonesian governing body for tech lmfao
failure
brought money from Australia
all you can eat buffet cost about $15 USD on average there
and you get to eat steak in there
The funny thing is, that steak meat is highly likely imported from AU 💀
Because domestic production can't keep up with the consumption
in AU you could definitely get a decent grade 1KG steak for about $14 USD
if you find the right one
though on average its much more expensive
Anyway, back to programming we go
a
i am not returning to indo aside from short visits
visual studio heavy af
steal his letter
I was programming in my dreams
i can find one under $8 

Got some bomb LiFePO4 for my project finally
Tactical nuke
btw, i encountered lags in windows notepad today
Yeah, but you should compare that with other GUI editor with lots of extension too
how? thats like the lightst windows program
Missed the news it seems
enlighten me
i opened big log file, and it froze
i blame the tabs.
Windows is turning to shit day by day
oh. big files it always choked on.
use Ultraedit or something which has a "large file mode"
and than i opened it in notepad++ without lags
Hello guys and gals, it's me Mutahar again! This time we take a look at what appears to be a situation where Microsoft pushed out an update that somehow caused the most basic applications to stop working... at all. How? Let's find out! Thanks for watching!
Like, Comment and Subscribe for more videos!
Crashes
i avoid win 11 like the plague
when i feel like using a gui editor i sometimes launch notepad from the command line.
but when wanna view large files in a gui i use UltraEdit because i paid for lifetime updates so damnit ima use them.
Its vast extensions and ecosystem makes it the worth
if you want get the open source version (code-oss)
Based

I think that is expected behaviour no?
If it's hibernate however...
(i did know)
"Sleep"
"Hibernate"
Is my pc a living thing?
why does it need to sleep
or hibernate
what? on linux it isnt, that actually stupid, if you turn of pc, why would you expect it to turns on just by touching keyboard
hibernate is a deep sleep mode, see?
hibernate != sleep
I always disable both
they are completely different
Which is why I make the distinction
My pc is on or off
its useful if you had laptops
to save up the battery when you're gone at the moment but wanting to use your laptop later
no, sleep for laptop is absolute shit
only hibernate
Hibernate is the one where you can't have input waking it up
or for PC if you want to
save up electricity bill
i haven't encountered any issues with it honestly
Because sleep is not turning it off. It just enters low power mode
exactly lol
10 months uptime, though never sleep nor hibernate
And that is what is actually happening on the CPU and ACPI too
"sleep" is just the screen turning off. There's a lot more CPU and motherboard power management stuff happening too, but you notice the monitor turning off
Read up the arch wiki for more info on the topic
electricity bill is questionable but im the same person who made an i9 13900 run at below 3W
and it is equal to turn off pc, when i use sleep, i probably not want it to be active in any way
just pressing anything right after sleep turn it back up
i had a dedicated display off on my laptop
it just turns the screen off but everything is still functional on the background
It is not equal in any way whatsoever
i suppose all 3 has use cases
Indeed it does. Sleep is when you're going to take a break for example
it all comes down to personal preferences basically
You still want the PC to go into the desktop
Hibernate is when you move place
And turn off when you're actually done using it
hibernate is to fake your long uptimes
no the point of hibernate is to preserve the workflow
To be fair, what windows call "shut down" when having fast boot is closer to hibernate lol
and sleep is the easy version of that but requires power
yeah. it's so you can put it to sleep while you travel or something
just turn it off smh
you guys don't shutdown your mobile devices when done using it?
I do lol
hibernate is purely software, i think... not sure
sleep really put cpu on sleep, it have much less actions happening during sleep
actually, probably it wait only on interrupts and nothing else happening
so it also dont require (that much) power or cooling
My laptop is currently off. I boot to arch faster than from hibernate anyway because no stuff to load from disk
Read up the arch wiki. It does provide a lot of info on what is sleep and hibernate is. Especially on linux
yeah but i want layout to be the same, i dont use tmux
do you have resume hook at the end or after.. autodetect? as i remember
Chat what's your last bios time
i'll go first
i have no idea
basically from bios -> desktop
that isnt only bios work
loader and os should do their work too
i don't know if linux had this kind of measurement built in though
there is iirc
quite sure it is lol
it is kernel lvl stuff
...that's what i said
what i mean is the process from bios to the desktop loading in
but if you have LUKS it'll count the time it takes to enter passphrase too
I have no clue how to find out
but bios have nothing to do with loader and os startup work
yeah the closest you have it systemd-analyze
charaverk | ~/Music/main ~$ systemd-analyze
Startup finished in 7.166s (firmware) + 6.924s (loader) + 1.464s (kernel) + 16.331s (userspace) = 31.886s
graphical.target reached after 16.317s in userspace
that isn't only bios work
7.1 seconds
i think
no wait that counts up all of the whole thing
yeah that what is firmware on motherboard
Something something funny stats
17
turns out windows does too
you should do something with it, or at least check is it ok for your motherboard
interesting kernel you using
I would really like to know why my kernel takes almost a minute to start up
what is your kernel?
Linux 6.14
general?
this is stats from work pc, i may used hibernation, but idk
Beeeg RAM maybe?
64GB, should be fine?
Idk, that's why I propose it lol
This system
couldnt make 40 sec difference
oh damn
so the rest of it was just you entering the password i assume
Or maybe the GPU is taking that long to boot up
yes
31 terabytes of storage 
Yes
Also I needs more
i waiting for a moment when new stuff will 'do not support x11'
Please keep supporting X11 thank you until Wayland is actually at feature parity and usable
That will not happen. X11 will still have support just like how tty modem still has support on linux
Wayland still lacks critical features like being able to remote into it remotely or being able to programmatically get the positions and boundaries of windows on the screens and then move them
that's like
most of the linux userbase that are using X11
it has matured way more than wayland
wait, but x11 doenst support that.. oh, no, it just have issues with load order, etc
as i remember it OK for wayland, dont know about moving windows, but hyperland runs on wayland, so window manager can do it?
Wayland mainly just sucks because it's GNOME dictated
I mean moving windows from an app like Neurolings
Which is currently just not possible to do in a way that works across all compositors
Wayland has a major fragmentation problem and major GNOME problem
being able to remote into it remotely
straight up misinformation, it supports that
programmatically get the positions and boundaries of windows on the screens and then move them
compositor-specific because it's intentionally not standardized, but implementing it for something like Neurolings is very possible if you just expose the right API (like wl_shimeji does)
does rustdesk support wayland properly yet
Wayland being a very low level protocol with a trash reference implementation was their biggest mistake
everyone out there reinventing the wheel and introducing unique flavors of bugs
who need remote connect to graphic shell now, for real
Well, I do sometimes need it to access higher performance computer at home
i thought the premise of AI is there's no need for it to sleep
servers down 
24/7 labor and all that
"The claude code doesn't need to sleep"
busy wait is the way
Neuro sleeps so idk what ur talking about
no she's just handcranking the generator to power the lavalamp
So that's why she's crying on latest paccha art
overworked 
Poor tony can't catch a break
Discord looks pretty graphical to me
so, use it, on your local device
I do everything on my own machine, I can't be bothered to set up Discord on my remote laptop
pretty sure you dont connect directly to your x server
what the fuck is wrong with you 
I like remote access stuff?
I've been using a remote connection setup for like 3 years now
First DRP for a while, then switching to VNC
Sponsor: Arctic Freezer 36 Air Cooler on Amazon https://geni.us/5VdU1
This benchmark of AMD's Ryzen 7 9850X3D CPU tests gaming performance, compares it against the best gaming CPUs out right now (like the 9800X3D, 9950X3D, and 7800X3D), and tests for power consumption, thermals, and frequency. We also tested the 9850X3D with some of the worst me...
it's a website
you can just open the website 
I can't be bothered to log in on my remote laptop and especially not use Discord without my plugins
if only there was a built-in way to synchronize all your settings too 
What does it matter anyway, I like managing my PC remotely so I will continue to do so
it can save space if you run it without a monitor. that's the only excuse i can think of
direct connection to graphical server slower than connection through app, with optimizations and another layer for stability at least, for simplicity at most (no need of any sort of user configuration for port for example)
using stuff that was designed in 1980(?) and used only up to 2010, is like...
just checked out x11vnc github, That the point lol
last change 9 month ago : text
prev change 2 years ago: actual maintaining
and other changes, countable per year
why dont you just use ssh fr
new best gaming CPU
I already do
But SSH can't do everything
SSH can't natively do GUI. so that's a solution.
could be it's perfect now.
direct connect to graphic server not sound like a good idea for me
It works perfectly fine though
vnc:
Well I tunnel it over SSH, obviously
by default
get a secure one
a secure one would not be vnc: protocol.
it would be a hack on top of a VNC server which required a special client
SSH has support for tunneling other network protocols over already trusted connections.
nothing special needed to make it secure.
I am still baffled at how amazing ssh is designed lol
SSH is peak
+4 FPS 
Let's face it, hardware has peaked. Now it's time for software optimization to be back
Huge
so my next computer will last a lifetime? i better spend big
TOFU isn't always great but can't really have a better default without telling everyone to just become a CA
generally a very nice protocol, tunneling + sftp are very convenient 
for something so widely used to have so few major vulnerabilities is impressive.
I mean how they anticipate future protocol expansion with ease and generally backward-forward compatibility between client-server
TOFU can always be overcome with side channel communication. Which, for your own machine is not a big deal
how the hell do i fix walls blocking objects with which the skyrim bot wants to interact
so sometimes it comes up to the thing but its behind the corner
idk i can probably make it interact through wall
or ignore it and send "cant reach" message
throw raytracing invisible ball to see if it collides with wall and try to wiggle
I see your problem. You're trying to make jank games acting logically
im using native skyrim pathfinding spell and its not giving super accurate results sometims
its still enough to walk around and pick flowers for example
imagine catching butterflies wouldnt that be cool
escape from prison should be good enough too
that would need some work
and fighting bot 
wait actually now i look at it, just walk, lock onto and smash the target
ranged weapons sound tricky though
it looks suprisingly good overall, no trace of shit happening in backend
the bot just walks where you told it and nothing seems to be off
it really looks similar to how minecraft bot walks
I wonder what geneva suggestion Neuro would break if she is integrated to skyrim
speedrunner strat
add ability to use cheats 
the more i look into it the more shit i see that will require a lot of work
for example, what do i do with puzzles
To be fair, games are famously not enterprise software
how do puzzles work?
like what does a player see vs what the API exposes?
i can get a list of interactive objects that fall into "skyrim puzzle" category like thosse rotating stones
i probably can get out their current states/positions
then....
i imagine some status updates will make it easier than others
give some kind of text hint on what to do with it
so its like
puzzle but textbased
not just answer but something more complex
"there is metal plate with fish on the right, bear on the left" bla bla
"move object1 to place2. move object 2 to place 1. move object 1 to object2." type thing
and then i tell it how keystones are aligned and they have to figure out
at least skyrim puzzles do not have much variety
that will save my ass
do you think they'll be easy enough to brute force without any clever tricks?
the most common one is 3 stones with 3 positions each
so just 9 combinations
however there is at least one puzzle that involves something more sophisticated
Lock pick?
4 stones, 1st rotates other 3, 2nd rotates other 2, 3rd rotates the last one
lockpicking is done (unless i will want to do some improvements)
currently i just ask for pick angle to try and it tries giving out how good it was
https://fixupx.com/testingcatalog/status/2016512802748309545?s=20
google will be.. collecting even more data 
BREAKING 🚨: Google is testing Voice Cloning on AI Studio, potentially ahead of the anticipated Gemini 3 Flash Native Audio upgrade.
︀︀
︀︀Users will be able to record and upload their own voices to let Gemini clone them.
︀︀
︀︀Soon? 👀
Yeah, soon finally every people will realize digital media can easily be faked. So hopefully it will increase people awareness
finally starting to see 1-bit/2-bit applications to SOTA OS models, god bless them for trying this out
(obviously, i wont be running that large of a model anytime soon, but a 60% decrease in parameters is insane if it keeps nearly same performance)
https://fixupx.com/UnslothAI/status/2016511345311834293?s=20
You can now run Kimi K2.5 locally! 🔥
︀︀
︀︀We shrank the 1T model to 240GB (-60%) via Dynamic 1-bit.
︀︀
︀︀Run at >40 tok/s on 240GB VRAM/RAM.
︀︀
︀︀2-bit is recommended as it passes our code tests. Run near full precision on 622GB.
︀︀
︀︀Guide: unsloth.ai/docs/models/kimi-k2.5
︀︀GGUF: huggingface.co/unsloth/Kimi-K2.5-GGUF
Quoting Kimi.ai (@Kimi_Moonshot)
︀
🥝 Meet Kimi K2.5, Open-Source Visual Agentic Intelligence.
︀︀
︀︀🔹 Global SOTA on Agentic Benchmarks: HLE full set (50.2%), BrowseComp (74.9%)
︀︀🔹 Open-source SOTA on Vision and Coding: MMMU Pro (78.5%), VideoMMMU (86.6%), SWE-bench Verified (76.8%)
︀︀🔹 Code with Taste: turn chats, images & videos into aesthetic websites with expressive motion.
︀︀🔹 Agent Swarm (Beta): self-directed agents working in parallel, at scale. Up to 100 sub-agents, 1,500 tool calls, 4.5× faster compared with single-agent setup.
︀︀-…
that is one chonker model damn
1TB.... That's the size of my whole laptop nvme ssd
unfortunately no amount of compression will help with a model as large as 1 trillion params
but it will actually ease up hardware requirements for providers with enough equipment
2TB bf16 
💀
all it takes is a single b300 for the 1-bit version 
1 trillion param will be higher than 1TB no? Because of the metadata and such
2 tb in total
oh also speaking of kimi k2.5, apparently kilo code made it free for a week if anyone wants to play around with it
So they compensated for lack of bits with huge parameters
kimi came out as any other model 16-bit with 1 trillion
unsloth compressed it to 2-bit
Such weirdly gigantic model
Who even needs models that big when a 30B runs perfectly well on a 3090 at an actually usable precision
considering that its said to be as good as 4.5 opus and gemini 3.0 pro and the rest of SOTA
all of those models likely have a gigantic amount of parameters
but unlike all those it's opensource
It's just brute-force
LLMs are at a cliffside that just keeps getting steeper
maybe, maybe not
Soon the cliffside will begin going backwards, if it hasn't already
depends on training methods that they use
atp everyone focuses on RL
let me guess, like vocal synth 
if there was something as good as LLMs and more efficient, they would have already done that lmao
replace LLMs with SLMs, trust
Vocal synth is cool, sure
But in this case I wasn't talking about that
diffusion models are a thing, but still ways off from LLM quality
gemini diffusion 2 when google
Just sharing my current hyperfixation on hyperdimensional vector-symbolic algebra:
https://www.hd-computing.com/home
http://gigasquidsoftware.com/blog/2022/12/31/vector-symbolic-architectures-in-clojure/
https://techxplore.com/news/2023-03-architecture-combines-deep-neural-networks.html
Using random 10k-element vectors of {-1,0,1} assigned to arbitrary concepts/data, one can bind concepts together (multiplication), bundle them up (addition) and iterate over them in sequence (permutation/rotation) all encoded in a single array. its incredibly exciting stuff
The most exciting for me is plugging this in to small-param (up to 30B) models. Assign a hypervector to each* embedding and use hypevector algebra select relevant embeddings and go from there
I've been working on my own ai today, was working fine earlier, now i keep getting an error 22, after trying numerous code changes, ive even tried old versions of my ai's code (which io know worked) and getting the same issue everytime. thus my ai's brain isn't receiving the input from the gui. Is there any resource I can look up that anyone knows of any chnages that occurred today? am in VS Code using python.
If you're using Python, you'd know if your code is causing the error by there being a stacktrace
If it's not that, it's likely something like a file permission error or just whatever OS you're using not knowing how to function and needing a reboot
gui not working might mean connection issue, depends on gui tbh
i would expect it to have a local-hosted web ui though, which means probably some sort of connection issue or nothing to register the gui input
if it's just in the terminal though it would be easy to debug
Thank you, rebooting has helped
hi konii
don't hit t
ogey
that would be mean
Hi konii 
I failed 2 of my exams 

My dude, both of those subject should be your expertise no?
Bachelor of Arts?
Game dev

And if there's one thing im unable to remember, its art
I dont give a fuck that johannes vermeer painted a cow
What about 3D? You delve into that before the game engine no?
Just lack of time
RIP
The people that passed have no life
I heard from my classmate he didnt do newyears or Christmas cuz he was working on 3D 
Wait the 3D exam is take home to make 3D stuff?
Am I overfitting?
Na it's probly fine
hello nerds

80% of the score is ye
anyways
the 140W usb-c arrived

i bought a 20 euro cable for it cuz they said it didnt come with a cabl
and it did come with a cable
so now i have 2 140W cables i guess
Those ugreen people wasting my money 
i should be able to charge my tablet at 45W now. before it would go down in battery while charging if the lcd brightness was too high
God what I would do for all of my classes to involve my major and not be low effort reruns of high school
gotta let those in rural areas catch up
Speaking of which, I got told that I'm in a rural area 
Sam can contest to how funny that phrase is

I don't think under any metric a city of this size is rural
when i think rural, i think of viillages
not when the population is 100k+
Sure if you count 10% of the major city as rural then go off ig, but I don't think that is correct
looks like fun subjects


It's a bat. or a curd.
does it want to hunt itself?
probably a bat. it looks like a flying mammal.
i made a cool thing:
https://github.com/alyfrillnya/NeuroSync

No git?
cool
You should look into photogrammetry next time 
do you think they wont notice the random 5 milion polygon model?
the whole point is good topology
and using trim textures and tiling textures
I don't know how easy it is between fixing topology of a scan or making one from scratch
a scan is close to worthless if you want good topology
When the vfx artist jumps into gamedev moment lol
I remember the legendary toothbrush from yandere dev 
So in the end you choose the computer vision way
if there is a light background behind the lamp, the lamp color itself will go white. it usually is shown with a black background but there are several scenes where it can be white sometimes. i think it adds the color to the background
“fixing topology” would be retopology in this case
which means
doing the whole thing yourself anyway

having the pointcloud from some photos would be helpful sometimes when you're making a new model but mostly to help you work out approximate dimensions. converting it directly into topology would suck and be bad.
huh
i swear ive heard that somewhere
Allegedly, the toothbrush model used in yandere simulator is very high in polygon count for what it is. It'a around 5k poly I believe?
Lmao, 1k poly is wack for a props
994 for a toothbrush is still way too much
Then do those acerola grass thingy?
3 if you're being fancy
no
already too much compute for a toothbrush
literally jsut a box with a 128x128 texture
for the bristles
the only justification for why it should be more detailed is if it's a first-person perspective and you put the toothbrush right up near the camera
That's some oddly specific thing in a game lol
it is true

belongs as a post on the blender subreddit
caption: toothbrush's perspective
hmm.. this model lets you average voices. i wonder if i could also do "make a voice that is the difference of voices A and B added to voice C."
(A-B)+C
kinda like embeddings with ("king" - "male" + "female") gives you something like "queen", would be interesting to see how it affects them

ooo.. i wonder what happens when you remove "the average off all british people" from a british person and add "the average of all american people"
Does anyone know how i can optimize Ny AI for 56 cores of my CPU? IT doesn't need to be Long Term Just for a few days BC the GPU IS coming afterwards
Assuming you use something like llama.cpp there should be nothing to do since they already have optimised simd kernels 
Just use llama.cpp and tell it how many cores to use
Also how big may the model be?
onnx might help I guess if it can fuse some stuff
Fr idk i mean i tried a few and every Model is okay for step one but there IS a Bit more
I will Look into that thx
The model size will determine if the GPU you're getting will be actually able to run it or you're stuck with CPU forever
what are you doing now? and what sort of models?
If it's a public model, which model is it?
Ik that but i meant for the Projekt and idk what model IT will BE (maybe 2 BC i need Something that is good with Databases
Well what GPU are you getting?
what do you have currently that you need to optimize? without knowing that, it's not possible to give meaningful or useful advice.
3060 with 12 GB vram
Well you're basically stuck with very small models
8B is basically optimal size for you
I mean ik what IT can handle ik what Models i will use my only question was If there IS any way how i can optimize the model on the CPU
optimize WHAT model?
No not really on CPU the best you can do is just give it all the cores
what are you doing now
Assuming you're using llama.cpp which is basically as optimized as it gets for CPU
coz you might already be doing it optimally.
are you actually running anything or just planning?
Yea i do that and other question when i use the CPU and gpu Like on the CPU IS the Database and everything that IT needs and on the GPU is the model would that be very slow or would it be fine to use
Databases are CPU tasks, LLMs are GPU tasks, that's a perfectly normal configuration
. . .
Communication between the CPU and GPU during LLM inference is minimal, so it won't be a bottleneck for database interaction
A GPU isn't really any good for running a database on either way, a DB on GPU would be very stupid
Ik but the Database IS Not a normal Database ITS a vector Database for the Model that is why i Had the question
Well I doubt it still makes any sense to put it on the GPU, especially with that little memory
I am running some Things
Yea that was my thought too
Assuming you are using llama.cpp for inference there are a few things you can try out to optimize performance, e.g. play around with the
- number of threads
--threads <n> - batch size
--batch-size <n> - context length
-c <n> - max number of generated tokens
-n <n>
not even going to ask why you only answered the least useful question.
and didn't say what you were running. someone else had to guess.
it helps to say "I'm trying XYZ with ABC and am having problems with JKL"
Sorry but i read what you Said
But XYZ and Abc a Not Something that makes a difference for jkl
Batch size and context length was nothing that i tried will do that thx
gpu-accelerated database engine 
btw once you get your GPU you are also not limited to 8b models, with options like flash attention, kv-cache quantization (and offloading of MoE layers for MoE models) you can easily run something like GPT-OSS-20b (just don't expect it to run fast when it does not fit into the GPU 😂 )
Just for fun i tried llama 70b and IT Runs okay on the CPU you have Like 20 min wait time but IT works AS Long AS you have enough RAM
240w or bust (i have yet to need full 240 from any usb c device except laptops)
20B, 30B, 70B, you can run it, as long as you have enough RAM
the brick is only 140W
F
im fine with 140w
i feel like it'd be harder to find 140w cables than 240
Has anyone tried making their own PCB with various method?
as in, making and etching at hoem? or ordering
I have access to toner transfer, msla 3d printer, and dry uv film with inkjet
At home
do the usagi electric special and cnc dummy thicc traces
I need smt with 2mm pitch. And I don't have access to cnc router/mill for now
next time i fly over mouser ill use my giant sky magnet and grab some
I don't want to try all 3 just to experiment that's why I ask here first in case someone else here has done it before lol
simply waste infinity money
im just using 240w cables
certified samvanmaele action to still refer to them as 140 because they are only using 140
i said the brick was 140w

new year new me
i was a different person yesterday
"Chad Seaman" what a name
he ain't afraid of drowning or sharks
what the hell am I looking at?
is this the average roblox game root directory?
Gotta try harder than that 
why are people pirating Xfer OTT lmao.. you can literally download it for free on the Xfer website since it's freeware
This looks like a beginner CTF challenge lmao
first line in the code and it already explodes because the attacker forgot to replace the placeholder string
Yeah [APP_PROXY_PATH] is clearly not base64 lmao
are they smart
no
what is that, like why?
is this deadass
bro got side quest
No that's the Roblox's engine source code (aka Client + Server Hoster source code
Yes it also includes the source code for studio
It's from 2016 btw
early 2016
surely it's been improved since then
Yep it was
There's also the 2012 source code out there
but it hasn't been leaked yet
And possibly 2020 and 2022
https://github.com/Artifaqt/ROBLOX2016 this is it here btw
Can be easily compiled if you have the contribs
I've compiled it one time
10 years ago
It has been recoded these days btw
a team of people
rewrote the entire client
and added new things to it
bethesda made actor bounds so big the pathfinder cannot find path through places where player can clearly walk through 
tried 3 ways to fix this and nothing works 
cant hook function that gets the bounds, cant get the bounds properties themselves
disaster
Does 2022 have their code obfuscation stuffs
What forbids you from the hooking? Function is smaller than 5 bytes needed for hook?
idk how it works, i have examples on how to hook virtual functions and it always works
but this one isnt virtual
its a function within some class
Say no more
https://github.com/cursey/safetyhook
actually there is instance where i hook non virtual function but it uses some kind of address that is being got with id
idk id for this one
or maybe i use it wrongly
Can’t you find it?
the library im using already has hook mechanisms
i just dont know how to use it in this case
i found something about the "cant get the bounds properties themselves" part
maybe this will work out
I can only assume the ID stuff has some ID to RVA mapping?
actually due to the commutative property you are like other girls
like other girls 
No bime? 
But vue is green 
Okay 
So the test line or train line would go up when it is classical overfitting?
Test line
Okay 
So I guess I can continue even when the curve looks like that as long both trend downwards ig
if you think about it recursion is like wearing a socks 
I had personal experience what Python thinks of recursion 
What?
oh i do
Every time the function is called aspy def main(): print("HAHAHAHAHAHAHA") main() main()could be described as if every time you call the funtion that you put on a pair of socks because Every time the function get's called, a return adress is needed for that which can't go on forever.
could be described as python sucks and doesn't do tail call optimisation
don't subtract Blue with Archive and add 24 on it 
It was just such a weird analogy lol
I know what tail call optimization is. But I don't see how it relates to socks
Long story short: a list that keeps track what function is getting called is getting overwhelmed
I could show it in a few seconds
Yeah, I know in high level how recursion is implemented on language like python
no wtf does recursion have to do with socks 

something i said
maybe
idk cuz recursions are layering architectures and socks are layering as well?
You layer socks?!
no zep so you're fine probably 
no I don't 
:mlntcandy:
Pretty much if that list is overwhelmed, the interpreter is gonna have a stroke:
@stone badge look at the confusion that your statement has caused
And yes, I didn't need to make that, but I felt like it
recursion doesn't have to be a circulated call inside one, or multiple functions
it can be just a chain/stack of functions that is not ending
Well, I know of that one...
Ah, that too 
socks kinda make sense, as you need to solve top layer to bottom layer

idk
i'm self-imagining to try to supplement this metaphor
True, And I do get what you mean
I can check how the call stack looks after some usage of my code 
whoever get confused has skill issues 
I don't ever layer socks 
you also live in hell tbf
Which is why I never need to layer clothes. If anything, I could do with less clothing sometimes
i only layer socks when wearing knitted socks from my grandma, which is to say never in the past decade or so 
That is mids init what the call stack looks like on my code... 
Toast, is that a simple call stack?

Sounds good to me 
wth im programming in C and just discovered C doesnt support function overloading
there's _Generic which sort of gives you this but manually
There is a way but you shouldn't do that
Yeah, but you probably should not tbh
alternatively just use c++, that's what i do
By python
@analog junco plz add support for this, accessibility thing, whatever it called
i like using this site, but having to grab mouse every time i want to list songs, or use touchpad - i not like it.
if contributions are welcome, plz give me link to github and i will figure out how to do it myself
probably html have special arguments/parameters/attributes, idk now 
also, if in playlist div i try to use scrolling, it just scroll down to that div, and doesnt notify* songs list, affect ability to use jk from vimium
i may formulate it badly, tell me if you dont understand i will record video
can you tell me a little about what this is? It’s my first time seeing it
BLT is crazy for text vector regression
wcag stuff
screen readers and whatnot want things wrapped in defined tags to programmatically allow them 2 support on the fly
aria tags etc
some people even fully abled will use these tools cuz they can make you into a wizard if you are good at them tbh
Oh i see 
biggest quickie fixiez just looking at the f12 worst offenders are buttons without text already part of them will just get called like
button
so you'd want to add aria-label attr to any of those
the random song acts like a button but isn't one by attributes rn so a wcag tool will just see it as the generic it is
Nobody:
My Humble Kingston Exodia (64 GB): Ventoy with Windows Embedded 7 POSREADY , Windows Embedded Industry 8.1, Windows 10 LTSC and Windows 11 LTSC. Also... Hiren's BootCD PE, Linux Mint 22.3 XFCE, Porteus XFCE (both 32-bit and 64-bit), rescuezilla, salix live 15.0 (32-bit), Ubuntu Server 24.04.3 and Memtest86+ (Both 32 and 64-bit versions)... And i still have 26 GB Free. (All Windows Versions are 64-bit only. Why? Because i think Win32 doesn't really matter that much, if you're daily-driving a Pentium 4, might as well use salix. Ubuntu server is there because who knows, maybe you're setting up an ubuntu server instance today, you never know in the world of IT)
...i may have over-prepared this USB
ideally each image should have alt="" attribute for screen readers, the worst one he's describing is definitely the button thing because random song is not focusable at all by wcag tools
it's currently in there as a generic (div/span) not a <button> or <a>
the setlist ideally would also be a list type
<ul>/<li> or role="list" /role="listitem"
gotcha. I didn’t account for those when working on it lol.
Thanks for explaining. I will look into it when i have time and see what i can do
i dont ever remember to think about it until i look at console in browser and accidentally open accessibility and see 
this is in console though, just look for accessibility and it will have a tree of aria attributes and stuff to make it easier
once you add them
since im already staring at it those button items, and alt text for images are ez but big
especially giving random a real div will instantly fix him being able to use it w/ keyboard in like 1 change
and then later on look at giving your nested <nav> divs aria tags to make it able to figure out what it's for, descriptions basically
oh and the play buttons on trending cards have no labels (as you've noted, they're all like this, but for quiccness fixing i'd go for the high traffic first obv)
got it
all those yellow label-y things in his screenshot are basically his tool going
what the FUGKK IS THIS TAG G
this one even done said ah fkkkkk
Those are the favorite, download and action buttons. It only shows up when you hover over a row tho
ah i didnt know (havent messed w/ the site)
it highlights them and everything because it juist reads the dom tree
it'll ignore them if they're labeled (unless appropriate for context)
it just sees the generic button div and shits them out because it's not sure what to do other than say yep this here is a button
learning something new today
this one should be more visible (in the overall mindshare sense) imo but it's very easy to gloss over
until you get the most heartbreaking message of all tiem like
"im a disabled certified hamster care expert and all my fingers have been chewed off over time, so i can't use your site as it appears there's no labels on fucking anything, thanks"
nobody remembers it exists until someone that needs it comes along and ...fortunately...? they're very rare
me included
i feel it's not very well disseminated info even though yes technically the standards exist
Lmaooo
Yep that’s true. First time i’d heard about it today 
always feels bad man to neglect them because theyre usually very sweet genuine people just unable to use the tools we take for granted like
a mouse
luckily it's easy stuff and generally just makes your life easier for maintenance purposes once theyre in anyways
Hello what happens here today?
i am a certified "why the hell do i need to put all these fucking tags bro i know what it is" enjoeyr
and then lose my mind when @nocturne olive asks for one(1) thing because i shortcutted 3 seconds 50 times and now i've got an avalanche of my own laziness overflowing
Wuh
Well anyway my network cable is absolutely cooked, I now need to reconnect and reset my network every 10 seconds with a script to have a half-functional internet
was talking about aria labels and div attributing and the existence of the accessibility tree/tools etc
sheesh
It was 60 before, now it's 10
i just made a few pull requests to the neuro sdk, yippee
also, waht
Script that makes my network actually work like half the time
;-;
My cable is definitely cooked

70 gigawatts of generation capacity are superbox-sinked from him keeping every single power state at wide open on every device on his system to patch jig it into giving the illusion of partially kinda working
billions of seals have cried
i guess 24 - 21 + 7
i don't really do the first part manually tho, i instantly think "that's 3"








