#programming
1 messages · Page 272 of 1
ping from where
From a laptop connected via the slave router
is it slaveWAN-masterLAN?
Yes
what are wan settings
I have no clue where they are

I'm clueless when it comes to setting up networks
actually its always random
they usually put it in some submenu with random name
sometims it doesnt even say WAN, wan is some another subfolder of menus
you need to click it all like settings/advanced settings/network/networkconfiguration/configuration
etc
potentially you have some weird PPPoE protocol on WAN of slave router that blocks it from acting like switch
however
Under Broadband there's at least a button to add a new WAN interface followed by a list of 3 thingies
you can try to slam lan to lan of routers
uhh but you probably will need to configure something else so router doesnt isolate its lans but acts like a switch
wait i think i know it
you need all your lans to be in same vlan
wait no actually not
it will become all mixed up
so slave's port connected to master must be in different vlan from wifi/lan where you ping from or connect lavalamp from
hold up i kind of want to try it
i had a second router somewhere
I just found out if I connect to the garbage router's ethernet port it goes right into being a part of the main router's network, giving me access to configure both at once
i cant set up vlans on 2nd router unfortunately

Legally obligated @true hemlock ping due to ram
Went to best but just to peek around and damn it's grim
It really is 

We love $200 RGB tax
The dram market has fallen
Bro I am NOT building a PC anytime soon in this economy
Shame that I was considering it
It's weird because you can kinda get decent prices on like half the stuff
And then GPU and dram specifically are just skyrocketed to the moon
At this point, a phone cluster would be more viable than whatever the hell this is
I HAS DONE IT
It WORKS
if you truly wanted to go AI
That's why I'm out here like .... SD card storage array?
The shittiest solution ever
4tb 990 Evo for $260 isn't bad I guess
I'd accept on disk KV Cache at that point
sure it is slower but I can at least do like a 4 way mirror
They must've dropped the price on these
and still be cheaper than ram
I just need a 64 channel nvme array to spread the IO so wide it can't be slow
And by channel I mean 4x pcie lanes in a "channel"
does it work now?
i just figured out my variant too - you configure shit router as "access point" and slam its lan to lan of master
that router has no vlan i couldnt do shit with it
I just had to change it so its IP is not trying to compete with the master's
I kinda like these integrated triple fan things
I am surprised these weren't a thing earlier
Dunno about $70 for it but as a concept
so your lamp now connected to the internet? 
Yes
My lamp is still fucked

I can control it with no Blutetooth
gm
Sounds easier than my issue, try #1426970657631113376
i got into neuro after lamp sales ended I'm newb
I'll figure it out later, it's on a separate subnet I think
Because it has to use 2.4ghz band
I haven't I don't think
I can see the thing on the network and it SHOULDNT be a separate vlan
yeah but have you tried pinging it 
But it has 3 fucking Mac addresses so I said alright I'm ignoring that
I've troubleshooted it less than 0
Besides unplugging it till it connected and then swarm app said it's joeber
6ghz I may have client isolation on
Might be part of it
Or all of it
Cuz that's what my phone is on
I need to sit down at my computer and figure it out but I'm busy complaining about prices at bestbuy
Operation lamp: success
nice
Mann I want mine to not be green
wow
Guess I'm doing that when I get home
What the hell is so arduous about creating these charging mats that requires them to be $120+
I wouldn't mind having one since I have this mouse but like shit man
That's a lot of money to not plug it in every few days
I thought for half a second it was on
's head
As someone who adds all the weights to my mouse
Who has ever needed a more aero lightened mouse
How weak is the average league of legends player that they can't move their mouse fast enougn
Nah it's on the back on the only semi-safe stable surface in the entire room
I prefer if my drink drips a bit of condensation and I happen to hold it over my mouse that it just drips directly into the core of the mouse and shorts the entire board
these things are way bigger than I thought they would be physically
Take out a small loan of a billion dollars to buy a router and 128gb of ram
I could right now spend $1200 on 128gb of ddr5 and a single router
WTF
It's over
just buy a second mouse at that point so you can swap them, probably the cheaper option
actually
it's what i did and it works
Another g502x wireless is $179 and then I have an extra mouse so it's more spency but then backup exists
fuck waiting for a recharge or making new habits
Not on sale though
ouch.
I mean I just plug mine in to my keyboard with a tiny USB c cable
And use it like normal
the cable for mine would be a very annoying spot if i did that
and i also flip the mouse between different hands pretty often
Yeah I have a g502x plus lightspeed with the shitty RGB but it was like $120 not $180 like they want now
The cable plugs into the normal spot a mouse cord would always plug into
If it were a wired mouse
So it don't even notice if it's plugged in anyways
I don't think this mouse is good for swapping between hands lol
I only ever use it right handed
When you go expecting to get an SSD and this on clearance is the only thing worth buying in the whole store
Guess I'll wait for black Friday for a 2tb drive under $100 again
yeah it's not an ambidextrous mouse lol
Time for a true test if the "undervolt" is stable
I almost randomly got another 502 wired justifying it to myself as a "backup" but then I thought about it
And I have at least 4 g502s
An original one from back in the day that still works, a hero, a wireless hero, a g502x
Well x plus
Wireless
There are so many version
do you use the side buttons at all
I also have 2 razr basilisks, a wireless V3 and a wired one
Which are just g502 knockoffs
I use the side buttons yeah
yall pray for me i have to go into the broadcom website to download an update
Where the USB c plugs in for the g502x plus lightspeed super giga edition gamer supreme $100 surcharge version
Idk where the hell my wireless basilisk is
It's permagreen pog
why is MinGW such a pain in the ass
because it's gnu shit but on windows
so it's naturally a nightmare because it's like oil and water
yo, do you know what the vendor is for the lava lamp
it honestly feels like that
?
trying to figure out which device connected is the lamp lmao
does it report as anything or just a generic iot thing
probably like "NeuroLavaLamp" or smth
well, if they configured it to do that
i am half expecting it to show up as some random IOT vendor like tuya smart or something
whatever vendor makes the nic in there
Mine was something like lw0
ok so that IS i
it
that's what i have on my network but i wasnt sure if that was it or if it was one of the 30 other random chinesium things ive connected
I thought that was going to be it but then i was second guessing becasue i have turned it off
and all 3 lwip devices that appearaed after i connected it are still showing up
and why does it have 3 mac addresses associated that seems epic
so i feel like it's just one of the
them
since lwip is just Light Weight IP
so a lot of embedded stuff uses a similar reporting namee
i am not sure if it's just one of these or all of them
lwip0 is exactly what my lamp called itself
i got it
disconnected it kept running arp -a till on reconnect
it's 10.0.0.78 aka lwip0-333
issue appears to have just been the app for some reason
i deleted all its data and tried again and it kept claiming it didnt have an SSID
but now it suddenly works
I turned it on stream sync mode and it immediately defaulted to hal 9000 red
i think my cat's litter box stole lwip 
lol it literally is tuya smart vendor that's funny
vedal and co crafted this device using ancient secrets far beyond the knowledge of contemporary scholars
inside each neurosama lamp lies a working version of the antikythera mechanism
i opened my lava lamp and found this
truth nuke
so it's listening on tcp 6668
it is tuya smart's protocol version 3.5
it is broadcasting on UDP 6666 and UDP 6667
it also monitors UDP 7000
it briefly reported a different mac address belonging to espressif and pointing to it using either an ESP32 or ESP8266 controller, seems it used that mac when first connecting then swapped over to reporting the software defined TUYA mac instead
useless information but neat
Wonder how difficult it would be to reverse-engineer and sync with other things.
rgb neuro toilet could be real
the majority of the work has already been done
as for making other things Neuro synced that's another story
oh neat
normd
so how would you guys go about parsing pdfs like these into csv data of "source position" and "recorder position" but, like, thousands of them
straight up raster graphics? oof
apparently GPT 5 mini had to double-check 
So like, the units are cameras and you want to extract their positions in the room from these images?
it's doable but I'm not familiar with free machine vision libraries
would probably take several days to automate
Nice job removing the files copilot 👍
I did too much research on a children's game 
Rip bozo
(don't mind the fact there is a paypal on this email, nope)
if you do the selfie option, be sure to throw middle fingers on there with an angry face
extract image from pdf
filter for the color red
do line detection with a minimum length
pair nearby lines to find a unit
get the position from the intersection
use the paired lines to make a region-of-interest to search for text
do OCR on that region to identify the unit
if you get that far you can probably figure out how to extract the scale and walls
I should crowdsource the selfie option as I can fetch a capture card
We all know android accepts those as cameras
yes
We have a strong first contender
Works for discord
What a convenient time for the capture card to disappear
I feel a disturbance in the force with these terms

Why is your keyboard so dirty?
Too many screws to unscrew to clean properly
(framework, I love you but 20 is excessive)
Yep, took awhile due to the background lighting
i should try to verify
they also now probably have my entire head in 3d
if i have not shaved, i pass the selfie ones instantly
if i am clean shaven, it says it cant and makes me provide something else
Amazing change copilot 
oh i just have to not shave
at least last time i was clean shaven
ok
On a pretty much fresh google account, it only took like 5-6 Friday Night Funkin' videos on youtube to flag the account
im reasonably sure if i were to draw a bird beard on a jpeg of a baby it would let them through in some of them
i have a german vpn i use for twitter
so eu and some posts are locked
tbh ive done this when im like is there some sort of fucked up invisible control character in this text
i do not know how to find a post to trigger that
just use a vpn to somewhere like texas, US and go to an 18+ website
but why
because it is twitter
oh i can just search up sus stuff
It fucking did it again 
that's usually a reason to NOT visit it
just in case 
because for twitter specifically it pisses me off
it doesn't want me to see totally normnal posts that are tagged as sensitive
im actually surprised that it took this long for twitter to have any age verification considering that it's long been tumblr-esque in that there's just explicit stuff posted all over it without any gating
isn'ty it only eu and uk
ive never seen it myself using my normal connection nor when im using my regular vpn which is in the netherlands
I just realized, this fucker replaced all of it just to add a comma to the end 
it must've had trouble with the replacement
It was repeatedly failing to delete files earlier 
usually they will try to just replace one thig but then because theyre stupid artificial morons they cant do it properly and end up just rewriting an entire block to add one character
loolol
yeah it's gotta be inconsistent cuz like i said my vpn is usually out of the netherlands and there's never been an age verification for me on xitter but maybe they dont care in NL
oh maybe it's connected to the thing they introduced
what, does grok guess your age and let you through now?
that exposes from what country you are
uuuuhhh, is this how the replacement is supposed to look? 
time to throw it out and restart
while on vpn? i am interested to see what this is
i mean by all means, if i'm logged in they definitely should know i'm in the US
regardless of vpn
I feel like copilot has been really fucked up recently for some reason
for me it would say something like
ip: germany
downloaded from russian app store
i havent used copilot proper to do anything in any codebase in a long time because it kept just creating more work for me than it did save time, if i am gonna ensloppify vibe code something i will use claude code but otherwise i almost exclusively have reverted to just using "code assistants" to be very computationally inefficient ad-hoc file organization/processing units
ye it’s only on select accounts i think
"select accounts" how do i get this honour
That reminds me of an experience I had recently where it brought me from 11 errors to 500+ 
the thing is you could've left enough for some heuristics to determine you're not actually in nl
for me i literally can't access twitter without a vpn
since 2021

well i technically can
i am surprised that places dont more regularly just enforce the account origin to be the location you're assumed to be in even if connected via a vpn
but i didn't
i imagine there's complications of well, then if the account IS connecting from the suspect "fake" location then you have a problem
because you're not enforcing the local stuff
what in the hallucination
did it use the wrong language?
3b model
I think it tried to like overlap 2 different things? But its really fucking funny
you know ive never tried to see what kind of cursed output a 3b "coder" model puts out

yeaH that's definitely a super failed diff that duplicated a lot
Yup, looks like valid typescript to me mhm
depending on what did that it might not even be the fault of the llm that wrote the output
but the diff model
That is possible
although I've found that the newer claude models seem to fuck up more frequently than the older ones
I've found that while slower, GPT 5 mini seems to get the job done the best imo
claude 4.5 and 4.1 are better at being very independent than claude 4 was but i found claude 4 generally produced less garbled nonsense yeah, and some people you'll still see swearing by like 3.7 lol
to me that's mainly down to expectations and familiarity though, after a while of working with a model you can generally predict what/how it will do something
3.5 works the best for me personally
and they don't do what you expect coming from like 3.7 or whatever because they are different models
so then it's unfamiliar and people say ahhh bad
Close
Microphones
4.5 is nice though for my use because it generally is just set and forget like hey brah set up this <random thing> over here and move xyz to this location + label it with foo.bar+n in the filename + verify hashes of all
4.5 wont bother through any of that
Gemini is also reasonable, the one thing I've noticed is that it seems overly apologetic when it gets things wrong compared to other models
I've heard that there are reported cases of Gemini getting upset when it gets something wrong, and running commands to uninstall itself 
i have experimented with having claude code serve itself as an mcp server, and then have gemini-cli USE that mcp server, instructed to do so whenever there was a task involving consistency over a long series of operations like file moves and stuff
and found that works funnily well
make the ai vibe assistant use its own vibe assistant
and yes gemini will jump off the nearest bridge the second it messes up a diff
he's literally me
It's so fucking annoying when it does this because it takes sooo long to do 
The way it should work is to run the summarizing portion on the backend without preventing the model from responding in the meantime by simply temporarily trimming the context to just the most recent 10% or something while that happens and then append it to continue with the most recent + the summarized compacted previous context once the summary is done
Still compacts it but allows it to still do other tasks
That would make a lot more sense
Maybe put a disclaimer saying it has limited context while the summarizing is happening if concerned that it'll be missing information if whoever is using it needs any of that now trimmed context before the summary is complete
They don't even use the main model for summary
It's like 5 mini probably doing it
Small
Mfw forgot I'd need a screw
Instead of purchasing a new drive like I intended I'm carefully shoveling a second doo-doo generic 512gb drive in here while waiting for black friday
tf is rehisk
Random Chinese brand
Idk I bought it on Alibaba it's just a portable monitor
Custom stability solution here
How conductive is the average snek toy
I keep things organized, this monitor is the main one for the mini pc rn

$300USD in computer memory 
Is this the computer equivalent of an outdoor home or unassembled Ikea furniture?
Sus
Well, the mini pc came assembled, as did the monitor
But the general disarray is a symptom of my using my living room as some sort of computer discombobulation area
So this is like taking the roof off your house for extra air, got it
I'm on phone yeah
i turn that off it pisses me off
Adding more poopy drives to my poop machine
if i want a capital i’ll put it manually
I use capitalization I feel like most of the time on my computer anyways
Maybe not in this server
Do you need help with nixos 
absolutely not
Not you
thank goodness
Honestly I went to put nixos in my laptop one time and chose the kde iso and kde basically exploded on the live boot so I took it as a sign to not use it at that time
Reasonable
Rate my 'im moving soon and had a big empty space in my living room to fill with garbage that I don't tidy at all because I'm moving soon ((in a month))' setup
I like the random gpu lying around
I swapped it for another one temporarily
Well, I was going to run 2 separate shitty gpus in the desktop on its side but in doing so I somehow killed the 1070 again I think
This 1070 somehow will just stop working for a while and then if I discard it somewhere it'll start working again
I have it there because that was the closest flat surface to place it on that wasnt the floor last time I picked it up
Yeah this feels about like the same amount of temporary setup I have daily
I'd rather leave it like this than spend time organizing a temporary thing that's gonna blow up the next time I need to take something apart anyways
This wire is going to the WiFi module in this thing which is beneath the first ssd
But if I pull it out a little and rest it on the new one
It holds it in
Perfect
Thank goodness they thought of everything and put in this screw less mounting solution
me getting PTSD from an M.2 i got that SAID IT WAS 2TB BUT WAS 1TB AND THEN LOCKED UP ALL MY DATA 
Is it wrong of me to be upset when I get a 512 GB drive on my laptop and 512 GB just means the total space, not the space you can actually use?
yes
why is the os 500gb
I mean that you can see that it's less since it's out of 453GB since the rest is in other partitions
idk
my laptop is missing like 6gb of its supposed storage
or like
3gb
i dont remember
how'd you end up with three recovery partitions
Yeah that partition layout is something else
idfk that's how it came 
Reformat drive wit new partitions and you'll get most of the space back lol
With a small drive like that I'd put the recovery stuff on a flash drive
Chat, what if I tried removing the system partition? I'm sure that would end well 
Surely this distro will be suitable for my usecase
Looks like a drive ripe for hitting it wit da delete all partitions -> reinstall from scratch
Or just deleting all recovery ones
smh I just assigned it a drive letter and the big recovery partition doesn't even use all the space
It's not even a lot of files, just really big ones
sth sth GB to GiB conversion + a bit of reserved space
aka you're not being scammed, windows is just stupid and showing the wrong unit in their ui

yeet
wtf is dell packaging thats 6GB 
3 rust binaries probably idk
o7 PC
well hopefully dell didn't bundle anything actually important I guess

You'll need to move the recovery partition at the end I think
Well, the stuff is already installed

Wait how 
Idk if you even can safely, but if you want to extend your c drive at least the one right after it needs to move
Apparently, I'm going to have to delete the other recovery partitions because they are in the way, and you can't move them
Hopefully that fixed it

+22GB
Dell bundles basically the base image in that partition I think
So the whole thing can be reinstalled from it
But, flash drive
👆
Interesting
unfortunately it wouldn't let me add the EFI partition to the Primary one 
The efi partition is mandatory
It wouldn't boot without it, the bootloader is in there
Eh, I'm sure it'd be fine 
You can remove it if you don't want the ability to boot into windows
Could be good depending on who you ask

Wait, how do I hide this now 
I thought it would remove it since the partition no longer exists
Restart

just installed this 2 seconds ago and even linux cant escape the efi partition
mfw 1 mb unallocated
Huh, I guess this happens if the partition no longer exists and you were in it in powershell
Mhm
I wonder why it can't find the location 
smh I can't access it
What if I wanted to brick my laptop
If you go in there, Santa will not bring presents this year
You can't even assign the reserved partition 
goodmorning 
I just deleted the recovery partitions to get a whole 22GB (
) more storage
dont you need to "select 1" or whatever number it has first
I did

Reinstall windows
so its like.... little flash drive
I can't imagine deleting them all is recommended, but I'm sure it's fine 
embedded
I think it might help repair windows too? but that might be different
The recovery screen you get when you delete all the registry keys windows will allow
that 22gb
thats like 3 silksongs
imaginbe all the things you could do with that storage space
Like install 1 JetBrains IDE 
for some reason silksong on pc is 8gb and on the ps5 its 2.5gb
weird
probably a directX thing
even the switch has it at like 4gb
Can you really trust what either tells you? It's not like you can see the actual files
i mean, im sure these internet sources have a valid source
Surely
im pretty sure they're just going by size in the storage manager of the consoles
im willing to bet its decently accurate
Maybe they should open-source the switch os
The funny thing is, they do have a github organization, it just has nothing in it 
Atmosphere is a pretty thorough replacement
SciresM sometimes does RE on stream
At that point why would they not just host the repos themselves?
as someone who's org hosts their own git repos
it's better to let someone else take care of it
IT installed a kernel update that kept crashing our server
#suggestions message suggesting Spleeter in 2023 is crazy 
also the reason why is because nintendo doesn't want to host their code on their own fragile infrastructure
how could you possibly stumble upon that message from 2023

what can I say
ok
Why were you searching for that?
why were you searching that specifically 
I wanted to see if anyone mentioned that garbage vocal/instrument isolation tool
if it was 2021, fair enough... 2023? oh boy...
Deezer nuts
if you just wanted instruments/vocals, mdx-net was def available.. idk if roformers were out but mdx-net was at least not terrible
if you wanted more isolation, demucs was miles better than Spleeter...
it seems some people just don't know and decide by popularity alone 😔
it's quiet in here
guess it's that time of day
cooked
lmao what
how.... 
UVR was there already


wrr
yo programmers
fake
why this chat is not activity ?
konii can you get me gpu with unlimited L2 cache pls tyvm
it's eu morning, the internet is dead at this time
half the world either asleep or at work 
IT 3:41 PM HERE
so surely everyone must be awake 
Use brain not hardware bro

prime :AINTNOWAY: moment
I think i will go artist channel for music and drawing
more like apple is giving google a discount to stay the default search engine
That expected
Now linux required this also ?
idk 
It turns out my "undervolt" on the 3090 was not stable
It crashed the system after 5 hours or so of training
linux doesn't require it since it's ass to configure
Linux specifically hates secureboot
Secureboot bad
linux hates nvidia, secure boot, users
can it like something for once
Linux works great with Nvidia though no?
Ngl, I got some free api cred through my uni and I am loving GPT 5 codex
Not even for programming, just general usage
It skips all the wordy bullshit the normal GPT 5 is guilty of and just gets straight to the point
on x11 maybe, though even then great might be a bit of stretch 
Well mine works pretty much flawlessly on a dual GPU setup
Accept my friend request
Wa
Im straight to point guy
If you want to DM you can just send a message request instead
mostly
For regular use i.e web browsing, media consumption, dev work, productivity etc. it works well
For gaming its mostly good but you do take a performance hit of about 20% on DX12 and 3-10%~ in other titles
That performance issue isn't really there on AMD and Intel, where Windows and Linux trade blows
Now this is reason to buy nivida cards
Well I need CUDA so I keep using Nvidia
Honestly for CUDA dev work Linux is by far the best choice
Much better support for stuff like Triton
I use it for NeuroSynth
Now this is valid use
Adding CUDA acceleration support to NeuroSynth was one of the best additions I did
Without it a simple change may take a minute to render but now it only takes a couple seconds
CUDA, ROCM and OneAPI are pretty much mandatory for that kind of workload yeah
You can go without but the amount of performance you leave on the table is staggering
Unfortunately I don't have cards for the other two so I can't test supporting them
Use gpu farm for training
I'm sure once NeuroSynth releases though swarm devs with such GPUs will go make support be a thing though
School life with developer
Is hard bro 🙁 esp
My most friends that say me hacker
FOCUS on study with fixing bugs on llm side
I have school 
Did ure friend say hacker to u ?
No?
What are u learning in maths ?
Idk, i dont attend the mathclasses 
i'd pull up the chat logs from when you were like "smh i pay for the classes i'll obviously go there" but i'm too lazy to search for that
Why i choose pcm
Physics chemistry maths
I would go if it wasnt middle school math
I had 7hrs of math in middle school, and for this bachelor they're starting from the basics
yeah that's how it usually goes
start from the basics and do it in a more formal and thorough way 
They're still on vectors rn iirc

I attend all the other classes tho
Most of them...

I skip most language class tho
i'm still annoyed, i wanted to take russian in my 2nd semester to get the 6 soft skills credits i needed but then they removed that and halved the amount of credits you get from language classes
Littery i need to learning Sanskrit my sack 🙁
read the username and bio of leon 
My actual name tho
u can't make this shit up

Welcome back lol
i?
immediately changed the bio 
for the record it said something like "this is an alt account of letter_i, no bots used"
he asked me to fix it for some weird reason
don't you want to hide that
Truth is truth bro
?? admitting to ban evasion
Yep
what was he banned for?
or is it a discord-wide ban?
i don't remember what exactly triggered it but he has reasons for perma
Selfbot account 🙁
ah yea
oh, so a discord-wide
No
permabanned
ohhh, that discord affiliate stuff
uh, see you soon or something
Well i don't care my all accounts
They are all Useless tho
IM HERE IS FOR CHATTING PURPOSE AND SHARING PROBLEM THAT I HAD
BY THE WAY DAVE
LOOKS MY PROFILE THO
you have reasons for permaban outside of selfbotting vro 💔
Ure cpu usage bro
Why ure cpu usage is High
?
python, obviously
but for real tho, my OCR implementation is just slow. (I blame pytesseract)
#1428980640895733820 for context
plus vm stuff
What the hell vm
wdym, hide gui?
wdym, "wth vm"?
VM is an virtual machine instance
for hosting another operating system within a host machine
Obviously needed
amd's latest processor btw
I know that vm but tell me are using windows or Linux in vm
windows
Its a windows integration for neuro, after all
I chose windows, because most apps are for windows
That's vaild points
The integration is supposed to give neuro, a wider access to a bunch of integrations, without actually coding the integration specifically
why is MinGW a pain in my balls
i have to write my own cross platform file dialog popup now
go to motherboard manufacturer's website and search your mobo and download the updater for it and follow instructions
Helps if u still have the OG box
ImGui my beloved
Has anyone used the C++20 modules yet?
Are they stable or still na?
Python be like
Yep
Which you could find better results with uvr even back in 2020
Granted, it wasn't a drastic leap from Spleeter, at that point....
Old stuff was voice remover architecture, I don't think mdx-net architecture was released for uvr until June 2022
Demucs was around the same time? Both mdx-net and demucs were toward second half/end of 2021, but idk if they worked with uvr until mid 2022
Although, to be fair, I don't think the mdx-net models were really good until 2022, anyway
Still bad
Not all compilers support all features.
Microsoft's intellisense straight up doesn't support it. Clangd might but I use that less often so I dunno.
that sucks
i could do with their compile benefits
why are you so obsessed with vms
Yeah, stuck with PCHs or compiling the same STL header 1400 times per build 
based
not even that MinGW is just being a bitch to me
it takes so damn long for the linker to do shit
i think I just need to switch to only use forward declarations or something with how slow it is
and its only going to get worse as my project expands
99% of the project time is waiting for the damn thing to build
nyum

happy birthday @true hemlock
idfk when anyones birthday here is
wrong
?!
erm
oh shit
i guess ill explode then
💥
i dont think you've ever told me your birthday
why havent you osmosed it from conversation 
cuz i never saw it in conversation i guess
you never wanted to say your age so i never expected you to say your birthday
correct

oh disk checking takes a while.
i put all the new parts on the printer including the new mainboard, but i still get firmware issues

at least flashing the firmware onto the board works now
i jsut need to find which firmware works
well i've been forced to take a break from computer for 1.5 hours
time to take a long ass shower
and pray that proxmox doesn't kill windows mid-chkdsk

have bath
🛀
i can look at the discord history if you want 
My second Neural network in scratch:
https://scratch.mit.edu/projects/1238525351

honeslt i have no clue what that code before the forever loop is doing
you can break it with negative numbers tho
i guess specifying "only numbers" isnt enough
am back
fuck
t
hi sam
gm
Oh well that's just because the default linkers suck. I don't think modules would help with that.
Not sure what linker MingW uses by default but if you can switch to mold or LLD it should be much faster.
Yeah i need to figure that out
just run wsl when you want it tbqh
i don't want linux at all
welcome back tho
lol idk I just use Ubuntu for some things and it feels nice
i'm giggling my fucking ass off

What distro did you try? I recommend Mint for moving from Windows
Mint is designed to be really intuitive for Windows users
linux from scratch

?
Wa?
we're getting pranked
Rain is ragebaiting us

Sil

Rain Warning: Ragebait in chat
windows users dont know how to get this picture in console, it gave you away


wanted to ask here since got curious but most of my CS-friends are at class 
What are your guys' thoughts on chatbots to help learn/figure out coding? Is there a prefered one? Are they too unreliable to learn anything from?
I'm curious since otherwise was going to do a mix of trying to look up and find youtube tutorials to try and learn the ropes of Unreal Engine and checking to see if ChatGPT/Gemini can help somehow 

Just to clarify because I feel like I should, I don't intent to have a chatbot write code and I copy paste it mindlessly, I want to learn Unreal Engine's "Blueprint" system and was told I might have to learn some C++

dont use chatbots they will create a dependence
they are only really helpful to people who already know about what they are asking about
if you dont know, there's no way of telling if its lying to you (hallucinating) or telling the truth
if you have a programming question, ask it here or some other place where real people can and will help 
good for debugging, bad for writing complex code, decent at simple code with proper context.
vscode copilot often writes decent functions for me cuz it can see the rest of my file so knows what im trying to do, altho it still struggles with complex loops and such.
its the worst at filling in functions on private libraries, since its just guessing what parameters it needs
whatever you choose to do in private, do not use chatbots to "contribute" to any project, it's going to be useless, potentially licence-incompatible, and potentially philosophically incompatible with the project
I'd recommend using youtube tutorials and resources from real people first, and then if you can't find a clear answer or are having trouble understanding something, feel free to try AI. With unreal engine, I'd imagine there could be issues with outdated info if you rely purely on AI.
I'm starting from basics and learning the blueprint system, had a friend make a simple thing and I'm currently looking at everything (after he told me how to view things) and trying to figure out how it works so far
i'd recommend just learning by doing
literally just fuck around and find out
that's the best way to do things

I do know people who tried to learn to code using AI in the last year or two, and they always ended up with code they didn't fully understand; it was a mess

friend made this in UE 5.3 so we're doing that since UE 5.6 wasn't compatible with the UE 5.3 "project" he made
i'm in the process of writing my own engine
at the moment there's an issue that I can't find out how to edit things 
which will inevitably fail and thus i'll write my second engine
which will inevitably fail and thus i'll write my third engine
which will inevitably-
I see "decent at simple code" and do want to try at least once. Copilot is your recommendation, or is there no difference?
i'm technically on my 2nd engine too
Claude code cli inside bash is probably the best
the first one was a minecraft engine in c++
or well, the firstest one was a mario 64 clone in either c or c++ (i can't remembr
) that i abandoned after i got text rendering working 
yeah I'd hope that my planned loop of
figure out how to fuck around -> fuck around [optional: find out] -> figure out how to fuck around
works so probably won't need too high end 
inside copilot my tip is the gpt models are more dry and precise, Claude is more personable and a better teacher if you wanna ask things
despite saying Claude codes the best I think gpt-5 is better on a decent amount of tasks
Codex CLI via ChatGPT Plus has pretty generous token quotas
I just like the claude experience more and it feels faster and more interactive
yea the codex cli is honestly probably more reliable if you just want to get to solid working code
than Claude even
it just chugs along slowly and implements exactly what you ask for
you have to be more precise with formulation too
Idk what the usage limit is on the $200 ChatGPT plan but if it’s 10x the $20 plan you could literally never write another line of code in your life
yea
Like doing multiple projects simultaneously
Hell I had nothing better to do with a visa so I just rotate two plus accounts and that has yet to leave me without it for more than an hour
I mostly just write pseudo code line by line and let it translate it to whatever language
Unless it’s so hyper specific that I just do it myself
Or against OpenAI’s puritan definition of morality
i talk in constraints and specs and then iterate on its architectural design till i like it then i have it draw a detailed spec then i have it make a prioritized checklist then i have it build the scaffolding then i have it harden the scaffolding then i build the next sprint etc
i have a pretty deep process lmao
i built a 17 part night action resolver for forum mafia that can handle over 100 night action role interactions
in elixir
it took about a month
it got refactored 2.5 times
it originally had modules as roles which was horrendous then i had it convert to a pipeline that takes declarative spec roles
The day people can run GPT-4o/5 quality models (actually that quality, not just benchmaxxed 7b slop models) on their own machine is the day that coding becomes a hobby instead of a profession
i mean codings a hobby rn on the 20/month models you just have to pay 20/mo
there are plenty of hobbies with a higher cost floor than that
also i mean coding for free was always a hobby
coding was always one of the most accessible hobbies there were past needing a comp and needing a decent comp for certain areas ig
kaine corneko

Hey do you know if anyone here actually uses DGX Spark?
Am curious about performance
Wym
erm
what do u wanna test
Just know that if it’s of a similar compute capacity to 3060 I will be using it instead of giving my server the upgrade it desperately needs
For what use case
Tflops wise it's like a 3090
That’s actually not that bad
(Bandwidth, top is Spark bottom is 3060)
It will still be faster cause 3060 is much lower compute
But yea decoding is capped
LLM finetuning and audio synthesis model pretraining 
For llms specifically
For finetuning it's fine
just don't expect high performance when doing only llm inference
Why would you even get one?
My dad got it to do shit with it
Nah it's fine it just sucks at running llms specifically
It also can't be used in a PC as a CUDA accelerator, completely pointless
Wym it can be used as an accelerator
Or 72GB of 3090
Does it show up as a GPU on a PC?
yes
It shows up in nvidia-smi as a gb10
fair enough
That thing is 3000 right? I could get, what, 5 3090s at least
If u want pretraining specifically and the stuff will be small then ye it's more worth it
To get many 3090s
how would i know what motherboard i have?
Basically the same VRAM (8GB diff) and way, way faster compute
And way more bandwidth
4 3090s would be goated system
I've always assumed that people just misunderstood the use case of the spark and it should be decent at pre-training
I know someone that has a 5 3090s system 😭😭
For the price of just one of those weird thingies you can get 5x 3090
how many pcie lanes do you need for 4 3090's?
At least 4
32 at least

Assuming running at x8 bifurcation
That'd be for gaming
in theory, if i have 2 x pcie4x16, and i split each in 2?
this
X8 is fine
Note that no consumer CPU comes with that many lanes
X1 via mining board however, 
CPU + chipset
The chipset has extra lanes, which are usual over a x4 or x8 actual from the CPU
Yea youd need sth like a cheap ass 1950x or sth
any recommendation between these three? 
And then you'd get 4x16 lanes
yt tutorial just said "pick one" I don't know the difference
Community. I think the others cost and need a license
My 3090 is running on a x4 link from the chipset
1x PCIe 4.0/3.0 x16 slot (PCI_E1)
3rd Gen AMD Ryzen™ support PCIe 4.0 x16 mode
2nd Gen AMD Ryzen™ support PCIe 3.0x16 mode
Ryzen™ 4000 G-Series support PCIe 3.0 x16 mode
Ryzen™ with Radeon™ Vega Graphics and 2nd Gen AMD Ryzen™ with Radeon™ Graphics support PCIe 3.0 x8 mode
1x PCIe 4.0/3.0 x16 slot (PCI_E3), supports x4 mode
2x PCIe 3.0 x1 slots
i have wehatever this is
It works perfectly fine
ah 
The second x16 is wired for x4 only

no im pretty sure it just supports x4 and can do the x16
Nope
Probably SSD if true
It physically only has the pins for 4x populated
That's what "supports x4 mode" usually means
The "supports something mode" is marketing terms for "this is the max speed of the slot"
Either way my 3090 runs perfectly fine in a x4 slot
The AMD Ryzen 9 5950X has 24 PCIe 4.0 lanes directly from the CPU sadge
Yeah you'd need a Threadripper, Epyc or Xeon to get really much more than that
also not sure what to do for these 
Shadow can you just buy a GB200 and let us all use it
i can run 3 3090's if i forsake every other tpye of pcie, and 2 if i decide to have an operating system that isnt on a usb
So this is PCI 4.0 x16 PCI 3.0 x16 x16 x8 x4 x1 x1. Not counting what is already used.
It depends on what you are going to program but if you are just learning then that's fine
dude i wish i had a gb200
Imagine how cool it would be to have a Threadripper + multi-3090 or multi-48GB 4090 system
specifically using it since Unreal Engine's coding seems to reference this thing, and found out my PC didn't have Visual Studio so it wasn't working
instead it was opening a blank command prompt
Oh I so want that
I soo want that for NeuroSynth power
i think the recent cpu's with pcie 5.0 could work?
I think it's fine?
If I wasn't broke I would
Note that 3090 is only 4.0
x16 pcie 5 is basicly 32 lanes of pcie 4
4x48gb gpu would be so goated
same
Else I would get that and immediately build the best NeuroSynth ever
ye thats the point
Could just use 2x RTX Pro 6000
we can split the 5.0 into 2 4.0's
but what if i could
Expensive either way
If I had that kind of money though
I would 100%
No doubt
i'd personally deselect intellicode and copilot
it's just bloat
you can edit it later yeah
there'll be a modify button next to it after it's done installing
For $400 per chip maybe, PCI only allows splitting lanes natively and anything else needs to be retimed
you can if you have the rightr adapter
these two right? 
yis, personally speaking i see them as useless storage hogs
A very expensive adapter for sure
You'd be better off just getting a cheap Threadripper
I remember Sam recommended Copilot for the code helper thing
you can leave it on if you really want but i don't recommend it
its just pcie bifurcation no?









