#programming
1 messages · Page 452 of 1
the gemma moe is also numerically worse than the qwen moe

the numbers are close enough for me to say they are roughly equivalent tho
if you were considering moving from Gemma 3 to Qwen 3.5, this might change your mind. But if you were already on Qwen 3.5, maybe not.
Here goes nothing
good luck 
Had to work out a few annoying little finicky issues with expected names being different from the documentation but it's going, and going surprisingly fast at that
ewe ubuntu
Its not the distro I use on my actual linux systems, it was just the first option for wsl which I don't really care about
I'm only using it for this one task anyway
i was (mostly) joking anyways. anyone using linux is a plus
I am rejigging this config again a little so trainings gonna take a bit longer
Tolerance is set way too high for my dataset
Bumping up batch size to 96 actually
Think I can get away with that
Or maybe even 128
Does kate editor from kde, supports ai
ew wsl
It's neovim worth it for i5 gen 2 2400, ram ddr3 12 gb, 300~ gb storage
Zed editor seems to be gpu based
(Gpu is a lot of money)
Very unlikely lol
If you want AI integration use VSCode. There is also a Claude integration.
no but it does have a pretty good LSP plugin system
Vscode is kinda heavy for me because I do a lot of things, so it's has a lot of extension for me, when I working, like AI, platformIO, C/C++, DevOps thingy, CI/CD, rojo, selene, luau, git extenisons.
So I prefer using neovim or kate editor on my lowend pc ig
then vim / neovim maybe?
Lemme see
also there is this:
https://github.com/yetone/avante.nvim
looks even better
i have no idea how u can use ai in vim
How does ai github account work.
I saw these bot account PR in a lot of github repos
Do you mean copilot ?
I don't see any of copilot bot account
You know you can also just tell Claude code or Codex to write commits for you
How? Do you message him and he PR you?
You can add the Claude GitHub integration if you want. Otherwise just tell it to commit for you. Just beware Claude loves to mention himself in the commit message. If you don't want that tell it not to do that lol.
My Baritone fork Great A’Tuin doing great
I move it to Java 21 changed the optional type and records, harded the code find about 12 bugs and a few say stuff like when player name = " Jerry256" do silly stuff
I let ChatGpt about 800 test write and find the invariant of each method in the codebase 80% done
Btw, I spent 20 minutes encoding that 
Idk why I did such a slow preset for that 
I mean, I don't mind spending 30-40 minutes on the under 10mb karaoke vids.. but.. that? 
Clearly you need to overclock your CPU
I used preset 1 svt-av1 and it's 1280x608
Despite looking way lower 
I intentionally starved as much bitrate as possible from the encoder
My cpu is ancient, though
3900x
Hem
I see
Sir if the 3900x is old and or slow then there is no hope for anyone still on am4
We should go back to AM3 for DDR3
Let's not. The bulldozer era chips objectively sucked. They sucked so bad AMD literally got sued over them
Hem
Simply make a modern AM3 chip using modern manufacturing, it can't be that bad
The issue was each core complex had a shared fp unit, so you had one fpu for every 2 cores
Simply don't do that
They did port a few of those chips to am4
They offered them as budget cpus on the first couple am4 gens
And some embedded opterons were ddr4
I had an FX-8350 back in the days
That thing was just an inefficient space heater lmao
Most of them were mine to begin with I let it optimize the PARALLEL THETA STAR and invent Adaptive ASTAR was the og implementation
I keep on running into this error a little ways into the training run
Can't seem to pin down what's causing it
either there is no data or it the wrong klind of data for this
Okay it was a torch issue https://cdn.7tv.app/emote/01GBPSCGR00007Q17796BDN5AJ/1x.avif
Switching to pyav fixed it
Its much slower but it works
Apparently torchcodec is quite brittle in its current state
We're in business
Failed to pass the first hurdle previously
Nice to know that they'll be removing the one that worked without issue for me soon 
There's an open git issue from a guy who suffered my exact error
Pytorch has still billions of open bug reports (means it is like super early beta)
Everything in ML is like this honestly, you run into so many weird bugs
To be fair they still find every few weeks things to make NN/LLMs way better.
I myself programmed new loose function which are way better in certain condition and same goes for activation functions which simulate hormones (useless in 3 bit weight environment)
I have one of those as a keychain, and another sitting in a motherboard in a drawer, the thing was hot as all hell, but it could play pubg at 40fps and those were the golden days
It was even able to play Tomb Raider AoD
Reddit still better than tiktok ngl
never a wrong decision
I should never opening a tiktok comments
what about the 3rd best time?
today but later
Ai hype started and tiktok filled with Ai doomposting or Ai ragebaiting
I don't see a single youtube comments behave like tiktok
that's like everywhere but i consistently see/hear people doing the dumbest things or having the worst takes imaginable and it almost always is tied to tiktok somehow
Should have never entered it in the first place lmao
it's specific to short form slop i think
like ive heard youtube shorts are godawful too compared to regular youtube
Agree

The worst thing is you might think that AI slop videos and images got less common
The reality is that it got so good that you just don't notice it anymore
just realized i got home and signed into my pc and am looking around like ??? kde is completely
lol
crazy take
Youtube compared to tiktok, youtube was less chaotic about Ai, yes the thumbnail are little bit clickbait like "Ai replacing all programmers" but the video wasn't about that
it really depends on what part of the website you are
but i feel like reddit is overall is still worse than tiktiok
no shot
i dont think there's a single piece of worthwhile content ever posted on tiktok

reddit has a subreddit for a lot of wild shit
what's wrong with that
Yeah let delete tiktok for a while, because it damaging mental health of our programmers
also reddit is so neutered now
actually the entire internet is extremely sanitized and neutered these days
cars fucking dragons
do you not see anything wrong with that?
at least the dragons fucking cars don't give people advice on how to soak their children's socks in pee with an onion or something and how that'll cure their measles
I only uses facebook
im willing to bet a lot of money there is a subreddit for that
Facebook seems to be calm, people are agreeing that Ai making software engineering easier
In large group of software engineering facebook group
there might be but it isnt getting 10 billion views like it does on tiktok since the way the site works is fundamentally different
Why do i know that sub
you can see it as sensorship, but you can also see it as good ridance
the whole internet in 2026 is this santized safe space in comparison to how it used to be
this place is like webkinz chat
there is a balance
old internet had some disgusting shit just for the sake of traumatizing people
still has
I checked X
Basically kinda same as facebook
Looks peaceful comments
Nobody cares about opinion
i havent seen anything that i would consider alarming on the internet in years
in terms of shock value
god damn it boomers they're the reason why we get shit like microslop
If you stay on the clearweb
Damn
HI
I think tiktok damaged my brain

most of the oldnet is on like i2p
if you have to ask yourself that question the answer is probably "yes"
I have returned to bug the people who know how to program about the feasability of making a neuro integration for an unreal engine game
We all should delete tiktok
So this is the first message I come to when opening a channel about programming
Neat
hang an onion on your belt
Put a zero next to it
WHAT
in my defense, i didnt bring it up
both c++ and lua are usable languages for this
for measles? yeah i find rust is the most blazingly fast way to clear it up
Sam, my brother in Muhammad, there are subreddit for the dream about wiseposting
People will it into reality lol

and both c++ and lua have SDKs! ok there's actually a shot that this can work
Alright done backreading
What SDK? Onion?
Meanwhile, vedal doesn't give a f, and still manually scripting in C#
While we are in chaos with Ai hype
who knows what's going on in neuro's cursed harness at this point
NeuroSDK is just websocket right? So more like API spec than an SDK
ionno
And Vedal will finally use TypeScript for the orchestration stuff
i'm super unfamiliar with it other than knowing it's a thing
im going back to basics, FORTRAN II only from here out
We have FORTRAN but no FORTWALK
FORTRAIN
FORTNIGHT 
Is robotic engineering worth it
I'm open source maintainer and about to major in this
it's a big field rn afaik? embodied ai is gonna be yuge
It depends on what you want to do
I'm gonna build ai humanoid robot girl
I'd say so
potato has been farting around with a robot arm
Then yeah, robotic engineering would help a lot
Someone has already does this, I'm gonna try it too
no wallet
no money
You can make a good and relatively cheap robot. The only bottleneck being the computing power
Maybe I'll just ask professor if he wanted to make project
So I can get free money
didnt someone here buy the roborock saros z70
with the robo arm
i wonder how that's goin
it's so silly
i love it
does it slap you so you clean your room?
Lmao
hopefully
gotta pay extra for yandere mode
dreame has a robo arm vac that they showed at CES but i dont think it can be purchased yet
that one they apparently are gonna make the arm also have vacuum capability at the end so it can like
dust

How long until someone brings one of those bad boys to the NHL?
thiccest puck ever
Close enough lol
i wanna see the national hvac robot league
there's been a few vaguely robot vacuum shaped robot wars bots
There was also one with a little pair of legs to climb stairs
Watch as they make one with both and we just get a weird little alien humanoid robot
This one lol
https://youtu.be/yqQ1KZ3Ox80
Matt Bores & Adam Wrigley from Bots FC are already known for building complex, exciting and World Champion robots, up next: Drift.
For their next project, codenamed: Project Stupid - they set out to put 250lb weapon motors inside a 30lb robot, taking advantage of the walker bonus which allows for 2x the weight - to do so.
Building the robot in...
most of the flagship bots can climb about 2inches at once or 3-4 with a slight stagger where it won't rise so i dont think they can do liek 4+ at once
like dreame's current x60 will do ~4.5cm twice
Nah I'm talking about real legs
is that the one with the giant saws
It hoists itself up, rests its head then wheels its legs up
lookin things
hell yeah
Another company also showed off this horrifically overengineered one
A couple years back actually
the motion of the ocean is pleasing
But the roborock one with wheels is way more elegant and actually viable for the market
the 5000 series is like a 25% performance boost and then X3D is like 15-20%
so... you can get like 45% performance boost on AM4 compared to what I have
holy mesmerizing
that’s made by a physics major not a robotics major
100%
it's so funny how complex the mechanisms end up for even then "basic" chassis lifts on the current bots
half the budget went to r&d, the other half went to whoever was figuring out that sweeping motion
Why do some c++ written by someone to control robot completely mostly hard-coded, I saw some 3000 lines of cpp in 1 file
cuz it was written by a masters student for a project and then got reused for eternity
Ig they don't care about software engineering
hmm
cleanest 2026 windows update
We love windows 💀
you may un-freeze it by preordering windows 12
so real
I guess I'll give it a bit longer
tbf I would say theres a 50/50 if u pull the plug on it it will just revert the update
Probably over 50% chance
I havent used windows in abt a year and a half but I did that a few time
Why the fuck are you gboard 
why is thanos
I'll give it 5 more minutes
Then it'll have been almost 15 min
You have more patience then I do
I would expect a percentage or something
I did not ask for gboard that isn't my default keyboard
windows major version updates like to take their sweet time
Precisely one of the many reasons I refuse to touch windows with a 6 foot pole.
But it's not installing the 25H2 thing
i touch it with a 10 foot pole sometimes on my laptop
Probably nuking the guys registry thats what mine loved doing
Still says updates are underway
Been over ten minutes 
booted that for the first time since feb a few days ago and it updated for well over an hour
It said estimated two minutes
yeah but the estimate is hallucinated by phi-3
that is infact accurate
(cant afford 4)
Okay, I think it's stuck
I'm unplugging 
As I will always say never touch microslop unless ur forced to by propitary software which wont run under a vm or you cant run one 🙂
if you quickly unplug the hard drive, crush it with a hammer just to be safe, adn then install linux on a new drive it'll fix it
I don't have a hard drive
Alright much better
man's running exclusively in ram
how tf did you revert
tbf u can do that on linux I highly doubt on windows tho
Lmao
I think the update finished
All it needed was a restart
But it was indeed stuck
do u use anything which exclusviely requires windows
Honestly I might consider trying the FUTO keyboard again because I'm already using it on my other devices due to Hackers no longer existing in the app store
iggly biggly i love u but i have no idea what ur talkin about rn
Eh, some applications would be more work with Linux
And gaming performance isn't fully there
thats is quite inaccurate actually
peformance is actually better on linux in most cases
gaming perf is better on my card running proton under windows games than it is running native keke
nvidia may be diff
let's not be delusional here, dx12 is still a big problem
I have nvidia
nevermind 
it is infact better
Privacy keyboard that happens to run a local LLM and is source available
iv ran dx12 games under proton like well?
Android virtual keyboard 
VKD3D is crazy
it's getting better tbf
I think what makes it better is that windows eats abt 5 gigs of ram after being debloated
so
Pc struggle
this one worked
lmk how it is, i remember last couple times I've tried it was just kinda bad 
tbf I understand why people use windows as a daily but I could never go back atp
I also tried on launch
iirc my biggest gripe with it was that the nn based one was pretty decent but it absolutely destroyed battery life
wonder if that's gotten better
been a while after all
tbf I might give it a shot if its half decent ngl
I been jus using the default keyboard in graphene
idek which one it is
might be the generic aosp one if i had to guess
so this is probs right
holy link
but the article is funny

defeated by outlook 
I'm glad NASA can't figure out how to make outlook work correctly either
makes me feel better
Microsoft seems to not be able to make software work competently anywhere
bro the amount of toilet related chatter on this mission
tbf thats every space mission
the toilet is a serious matter
yeah
i just havent like sat listening to the comms on one for a hwile
ye fe
what else is new
I nomally if I look, end up reading transcripts
not much just funny that there even using outlook on a space craft
I would of figured getting something custom made up would be more logical
since outlook is made by microsoft and hence is inefffecient and a buggy mess

outlook used to be much better when it was a native app
ye
actually the more I think about it the more I remember that electron isnt bad its just so unnecessary when microsoft uses it
ye, u can actually make electron run reasonbly well
I have done so in the past
its quite handy for just spinning something up, but microsoft seems incapable of doing any basic fundimentally to make it run at all well
as evident by obsidian
ye
I personally didn't go with it because of lack of keys like ctrl that hackers keyboard does
u get ctrl key?
I have all of them 
No longer exists on play store tho
tbf I use the play store for the bare minimum anyways
Like it doesn't exist anymore period replaced with slop
I have it installed to get some apps I have previously paid for on there
thats abt it
if I didnt want those I wouldnt install it , one of the benefits of grapheneos
If I wanted to share I'd have to yoink the APK off of my device to share
fe
I can probs hunt it down
theres an archive of all removed stuff of Google play
I just needa grab the link
anyways I gotta head out its like 1am and my sleep schedule is already an abomination
HEALTHY GRADIENTS AND LOSS 🥳
LETS GO
We're on track for a successful SmolVLA training run 
Ironically the bottleneck in training speed right now is PyAV and not my GPU 
Lmao i just got around to watching yesterday's stream.
Has Evil infected Neuro now with the fim_suffix token virus? 
<|fim_suffix|>
<|fim_suffix|>
NASA’s mission to orbit the Moon is being interrupted by Outlook (New) and Outlook (classic) both refusing to open 😂
"FUCK MICROSOFT"
That's a joke right?
Must be late April fools
Outlook is yes
The Windows experience
Simply install Linux


Why doesn't my distro get a cute anime waifu mascot 
be the change you want to see
yes
buffering?
which distro
i mean it has to buffer and cache all the video data
It's 2026
Jarvis, concentrate all your power making no mistakes to produce a production ready lossless video streaming codec that uses no memory or bandwidth
on it boss
ive done it
what the hell, it's noisy
its done

Seems pretty legit if these benchmarks are to be believed
Art
headache attempting to get the supported 280hz actually working with amdgpu
finally got a working edid
up from 240hz at last
@tender river look at this
amd gpu
Simply have a monitor that caps at 240 like me and you don't have this issue
I have thought about getting a 1080p with way higher refresh, is it that worth
Ye, I don't. Have any monitors under 2k rn so I'm scared of going back to 1080 ig lol
imo oled with lower refresh has better motion clarity than ips with higher refresh
What's the g2g response on that
not sure
my monitor is vg259qm
rtings says "exceptionally low response time"
Lol my vertical side monitor is a vg279qr
I didn't realize qm was 280hz
QR is 165

lol
https://forums.blurbusters.com/viewtopic.php?t=7414
i followed this to enable od120 on my monitor
whatever this means
holy overshoot error
tentatively I'm seeing up to od80 on the 279qr
yes, you need to enter the service menu and enable od strong-weak
136.6% is bad, yeah
What's overshoot look like in practice
reverse ghosting
Gotcha
its really not visible for me
Blindge benefit
the ghosting is pretty noticeable without shadow boost or with od100
I'm about to find out if I can see and/or care about it
od80 is better than od100 for the monitor but od120 looks better still
Surely maxxing the monitor almost double usual refresh has no issues or long term concerns

surely
to make sure, you know od isnt overclock right 
my monitor only overclocks from 240 to 280 by stock settings
Mine after I've been ignorant of this for like 5 years+ of having it suddenly getting turnt the fuck up
Better use of it than the text and YouTube videos mine will see
im nearly moved to tears
280 looks so good
just played osu with it for the first time in 2 years
Fuckkkkkkkk the buttons are on the other side
Minor inconvenience it's over
Oh they're on top
We're so back
lol
Skill issue
I just wish the OLED monitors weren't as gimped as they are
the G5 can hit 2400 nits on a 10% window... but a lot of the monitors are limited to about 500 
G6 may even be close to 3200 nits on 2% window.. but not sure, HDTVTest said there were issues so gotta wait until those got smoothed out to really know
my alienware is qdoled and it can reach like 900-1000
in hdr
idk what would need more than that ngl i keep it at like 50%
maybe if your environment is bright asf
game optimizer mode is usually lower.. so it could just be that
but the PQ EOTF tracking is way off so can't trust these brightness measurements
gotta wait 
going for good black levels hurts the peak nits itll do as well
I'm also wondering how much improved anti glare the G6 has
the G5 was already amazing
C5 versus G5, lol

goodmorning
true
i got awoken by construction in-front of my house 
rtings now has a membership i need to have to see how my monitors compare
supposedly mine have this
idk what they mean with fps limit, these are 180hz
Ad revenue from text-based journalism has been collapsing and LLM's pulling article summaries into search results is accelerating the problem.
that's fair i guess
Paywall is a desparation move and I would not be surprised if the site shuts down within the year 
i dont really want to get a membership for a website i use once a year tho
rtings membership is 45 bucks per year
i feel like this graph is without the response time on "fast"
my monitors should be pretty good for response time iirc
i bought them exactly for that reason
yeah, they were going to put ads into rtings but there was a lot of backlash and the amount of money they would have gained would have absolutely not been worth it
I do think they expanded a bit quickly and that's part of it, but, eh... probably was inevitable nonetheless
yeah, it is $45 a year.. not too much.. the issue is a lot of people just want to look at one review every like 6 months
and they charge $10 for a month which is a little steep
To be fair, I understood the need for money because hosting and the journalists costs money. I'd gladly pay money if it means I got quality journalism
And ads just don't cut it

response time isn't by any means bad.. but not the fastest IPS panels, either
but pretty close
the overshoot error there is very low
the overshoot gets a lot worse at "fastest"
i feel like "fast" is the only usable mode
I do wish they put more tunable settings
Net positive for humanity
These guys don’t think information should be free
Smh
if it was free then why would people spend the time making it in the first place
Information, not art
Big difference
not really
eeee
we pay reporters to go outside and gather information.
so we should also pay the people that put the effort into testing stuff and reporting their findings
Sure information should be free, but journalist also needs to live somehow
there is a reason information is seen as valuable
I believe that information being free and human needs to make a living should not be a mutually exclusive goal anyway. I just don't know how, but maybe someone smarter than me would figure it out

ya go out there work unpaid mate and see how it feels
llm just sorta ruins it because not only you get shitty ass hallucinated summaries
instead of those people who provided information for free but still living off ads now they had to paywall everything
made things worse
also hardware analysis and testing ain't cheap we also need budget to do that shit.
some of us that are knowledgeable enough to do complicated tunings do not have infinite budget to benchmark every single bga soldered cpu modt combo out there so once we bought them and benched them we'll need to sell them quick for lower price

honestly should've been grateful that there are people out there going as far as buying hundreds of sd cards for analytical purposes and gives out the data for free
and you still had the audacity to thank the llm instead for stealing their work
We are programming human now
programming
gramproing
@patent shard i haven't tried it

yippie
https://huggingface.co/GaboxR67/MelBandRoformers/tree/main/melbandroformers/experimental I say fv7beta2 because beta1 and beta3 are even fuller/noisier
voc?
yeah
ohh ye me stream osu earlier
union dt pass
oroka dt1.3 nearly pass
https://ordr.issou.best/watch/ZXj03b <- oroka
https://ordr.issou.best/watch/FAbeun <- union
where the configs?
onion
only thing is it does have slightly more instrumental bleed than some models
uh...
oh, right
can i use the one for the inst model?
wait
wrong one
https://huggingface.co/GaboxR67/MelBandRoformers/blob/main/melbandroformers/vocals/voc_gabox.yaml this one
i love githubs view raw. then i only need to hit ctrl s
can't you curl the .diff/.patch file?
why for a new file ive never used befir? besides git has it's own internal diff tooling
@patent shardthank god i didnt have to tweak any config files this time
RIIGHHTTTTT the UVR linux build isn't 5000 series compatable
maybe try MSST
the dev is working on an update for uvr but they said there's a lot of tech debt
oh I'm fine. I just foced it to run on my 3060
i wanted to use my 5060 ti since ti's faster
it takes just a little over 10 seconds to process a song for me on 4090
I guess that is a lot faster than 3060 😅
on a lot of stuff i have to use cuda 12.9
most software is tolerant of upgrading the deps thankfully
yeah a about 30% faster, but more bleed than using the INST model
there's def some more bleedless models
big beta 6x is still pretty good.. it is older and doesn't capture harmonies well, though
there's also fv7 gabox.. I found some issues with it, though.. but maybe worth a shot
https://huggingface.co/GaboxR67/MelBandRoformers/tree/main/melbandroformers/vocals fv7 here, and v7.yaml
big beta 6x here https://huggingface.co/pcunwa/Mel-Band-Roformer-big/tree/main
i tried the inst v7 i think, but liked 10 better
doesn't always capture them well*
inst v7 is prob more bleedless.. and instrumental models aren't really made for vocals
so the fuller v10 probably worked better, yeah
in my case i normally only need the voal for transciption anyways
in this case I'm building a model
normally i need a good inst for final mix as well so garboc flowers v10 woorks perfect for my normal use case
I like deux.. but it does have some issues
some songs just come out really supressed and are lacking fullness.. and there's vocal bleed during quieter parts that unwa models don't have
deux is less noisy
but it's all about trading fullness and noisiness
hoestly I know it's not ideal, but I like the balance v10 gives me RN
what I'm really hunting for is a good local backing/lead model. I like the MVsep kara model, but sadly its on MVsep lmao
frazer & becruily and anvuew are good ones
gabox also made one.. haven't tested it too much but it can also be good
really depends on the song.. anvuew is generally better for lead vocals than frazer & becruily
anvuew is more aggressive
and you generally want to run it on the full song, with the instruments still in there
because otherwise it tends to get confused easily if you extract the vocals first
aah
Btw how many people have already received their subathon duality shirts?
I'm still waiting for mine. It hasn't even shipped yet :/
Oh well the email i got after i ordered it back in December said "Ships by April 30th"
Guess i still gotta wait a bit
hi
@sage crag
hi
hi

?? what did i come back to lol

@amber fractal

You can't do anything while I'm yellow
roll down hill
im jsut matching the energy
hm
5 booms
this is so silly blep :p
cant unyellow
peak #programming
happy birthday
dappy hirthday

bippy harthday
it's still 110 days till my birthday, but i dont mind konii being early
who bippy
kaboo m


👀

I heard claude tracks swear words usage
actually why did I post that screenshot 
your neurocord is so cut down
that it is
I wonder if claude also tracks the use of the word clanker
OM
aintnoway genchat or the neurons does not exist for you 😭
I did edit this but now I want to know if I can actually make it permanant
you're gonna have to use css mods for that
Compact mode shows times, but I removed them
mainly because I don't backread on PC
much
Wow true dark theme
I thought most people just use second dark
It's call ash?
should have added "Ultra" in the beginning to make it ULTRAEPIC MEGAGAMES 
wait i forgot to also add "pro max plus"
there's "ash" for gray, "dark" for dark gray, and "onyx" for nearly black
Sorry second dark sounds stupid 
nah its fine lol
working on it, I'll do a botch to start then make sure it only targets neurocord later
we love the CSS content element to rewrite the content of an element
nope, not quite
now somehow center aligned
not center aligned, just existing in the document order but not showing
discord css is kinda hard to edit ngl
it really is
i mean even through just inspect element without permanent modifications its a mess of a client

i kinda wonder what other web apps (eww) could even compete with discord's crazy ass dynamic html handling
/* Remove, then replace */
a[href="/channels/574720535888396288/1071784467036913664"] div.name__2ea32.overflow_b0dfc2 {
text-indent: -9999px;
line-height: 0;
visibility: hidden;
}
a[href="/channels/574720535888396288/1071784467036913664"] div.name__2ea32.overflow_b0dfc2::after {
visibility: visible;
display: block;
line-height: initial;
text-indent: 0;
content: "autismu";
}
let me actually update that rq, there is an a tag I can fetch channel from
gonna use this for gymbag general lol
guys im gonna try to invade every single website iu can find tonight
any tips and tricks chat
let me edit this to be more user friendly them
because the a tag targeting is a lot easier to manage

you're gonna do what?
Let me see if I can do channel description as well
classic microsoft https://fixupx.com/SpaceKoala/status/2039874363185660043?s=20
Inspiring
Quoting Grant Tremblay (@astrogrant)
︀
“I have two versions of outlook and neither of them are working” is actually a generational NASA quote now. Not quite One Small Step but every generation lives in a different world
this is the one step for mankind typa shit
why did both options fail
this is a grave issue
redundancy is important in spaceflight
i demand a nasa investigation
guys i larped programming and recorded me booting up nodejs on instagram, and a girl told me im her type
I wonder why they didn't make outlook a PWA
ok
someone give me more things to download so i can larp
is arch linux too hardcore or could you turn it into larp material
I'd not recommend yoinking yet as I am going after the one on top and the desccription still
but tbf it is already useful as is
the top part is going to be a pain because I have exactly zero identifiers in this stack for what channel the iser is on
mabe due to the botch to hide it? this method essently places the custom text under the orginal and just makes the orginal have zero height

should be in block mode so might be possible to offset
arch is definitely up there on the larp scale
I am not seeing a method here, nothing in the node tree works; only thing that maybe could work is seeing if the selected channel is active but I don't see a way to detect classes on other elements
typing "arch" into the gif finder was not what i expected
i forget stuff other than arch linux exists

for completeness sake, I'll just make a thing that does it regardless for if a botch is found
1.5 line height on the replacement text looks correct... 
💊
You make me search it...
/* Remove, then replace */
a[href="/channels/574720535888396288/1071784467036913664"] div.name__2ea32.overflow_b0dfc2 {
text-indent: -9999px;
line-height: 0;
visibility: hidden;
}
a[href="/channels/574720535888396288/1071784467036913664"] div.name__2ea32.overflow_b0dfc2::after {
visibility: visible;
display: block;
line-height: initial;
text-indent: 0;
content: "autismu";
}
/* always active, requires botch still */
section.title_f75fb0 h1 {
text-indent: -9999px;
line-height: 0;
visibility: hidden;
}
section.title_f75fb0 h1::after {
visibility: visible;
display: block;
line-height: initial;
text-indent: 0;
content: "autismu";
}
section.title_f75fb0 div.topic__6ec1a span {
text-indent: -9999px;
line-height: 0;
visibility: hidden;
}
/* tired, cba tonight to fix */
section.title_f75fb0 div.topic__6ec1a span::after {
visibility: visible;
display: block;
line-height: initial;
text-indent: 0;
content: "autismu";
}
The only botch I can see if the a tags container has the active class
hello chat
hello nonas
i see
anyways i am planning on making a clank— i mean llm
should i use pytorch, tensorflow, or something else
PyTorch is recommended for newcomers
alr thx
clank is my favorite compiler 🥰
Translate C to LLVM IR via LLM
“Don’t, use Clang” or “Don’t use Clang”?
Well you should clearly be using MSVC
It is

cheesi
konii
Maybe there is some internal Linux version for Microsoft internal CI or sth
ying 
look what i went through to get 280hz working on my monitor, which incorrectly reports its limitation
konii? ying?
choose
modelines 
kill
who is going to be confused
idk
not me
the top one is the one that's supposed to be the edid for 280hz, but it brokege for me
so i fiddled with the numbers until it worked
they are
shouldn't be necessary anymore since VGA is dead
welp
-# mostly
iirc DVI still gets used
https://www.reddit.com/r/linux_gaming/comments/19a50dd/asus_tuf_vg259qm_wont_run_at_over_240hz/
everyone has this issue, but im the one who has it in a nix config 
I used DVI until a while ago
but new graphics card lacks DVI port I think
hdmi has dvi support
well, a lot of the time it does
the lightcrafter 4500 jsut uses dvi signalling through the mini-hdmi connector
Oh, is it arrived yet?
it arrived ages ago lol
has the magic smoke issue been resolved yet

im gonna buy a new one
but i currently have 1K spent on a gaming pc, and 640 in a 7900xtx
the 7900xtx seems to be legit, and ill sell superbox my 3090 once it arrives
god. 9070xt feels so affordable
it is
i could probably have gottena brand new 9070xt for how much i spent on that 7900xtx
i'll probably get one paired also with my 4070 super once i got my ram back and build a new pc
but i feel like for 4K gaming with the number of dumb mods we'll be putting on triple A titles, the 24gb of vram might be a good idea
and the raster is just better
the 1K pc im not sure if ill be seeing it tbh
this is the 2nd attempt at buying it after getting automatic refund before
idk what the seller is doing
god fuck
already spending thousands for my new briefcase
then i immediately went straight to building a new pc for myself
Heads up, Gemma 4 was broken on llama.cpp last night
Apparently they've merged in the fixes now
what exactly was broken about it?
^ in case you were testing
The usual, tokenizer bugs degrading performance
Broken tool calls too
Wait, he burned it? 
Ohh. How?
if i killed the projector id be out 1K, the pi was only 280

either shitty buck converter, ground loop, or 5V shorting with the 3.3V line on the pi
Oh, and the quants are broken as well
My guess is the last one. Because a shitty buck doesn't usually result in something burning
ye, I thought the same
its the power contorller thing ont he pi that got cooked
one of the inductors next to it mostly
none of it has markigns of what i need to replace it, nor do i have any clue what is and isn't broken
so ill just have to get a new one
shitty buck converter would mean bad power which may result in the Pi not working properly, but shouldn't fry anything
ground loop should at most be a signaling issue between the Pi and other stuff, it shouldn't have enough current to fry anything (and I don't see how it would happen, given that everything is connected to the same ground and is right next to each other)
the motor controller's 12V and 5V stuff has connected ground
apparently
well ye
That's normal
there's no reason to have separate ground AFAIK
Ground is a fancy way to say 0V reference. So 5V and 12V on different GND is in fact may be the same
the electricity from the motor could have gone through the pi instead of through the fuse
your motor driver should make that impossible
That would depends on how you wire it. But I don't think it's the issue. Even if you have a spike from the motor, the driver is the one that should take the brunt of it
i dont trust that motor driver for shit anymore
im not powering it from the pi anymore, ill give it a separate buck converter
doesnt need more than 1A anyways
its def a short. shitty buck still can't supply enough current
That's what hsinu said no?
at most you get weird transients/ripples from shitty bucks and/or some weird emi
i put resistors on thew data lines. is that not enough?
note that you definitely still need to connect ground between the motor driver and the Pi, so idk what using a different buck converter accomplishes
Not really protects from voltage spike
the motor one was 12V 5A
no

5A
honestly thought 280 for a pi is... what the fuck

Orange Pi 5 mind you
i need one that would do 570x1140@180hz
shouldn't all the buck converters just bridge input and output ground anyway
You may also want to try FPGA 
no
Most bucks do, yeah
But there are some that isolate too
so
However, the current rating is usually shit
if i still power the motor controller from the pi, is there even any need for opto-isolator?
im using the 5V from the gpio to power it
That... is not ideal
no
See?
it takes only 5V for the control stuff
for the motor itself it takes anywherre up to 24V or so
Can you send rough sketch of how you wire it up?
sure
not my drawing but pretty much how id use it
the motor controller has these "R_EN" "L_EN" pins that need to be high for it to want to even power the motor
i think i might have accidentally connected those to 3.3V while powering the VCC and GND with 5V
my best geuss is that fried it
ye, I think that would do it
also powering both R_EN and L_EN at the same time mabe
(if you did that)
powering R_EN and L_EN at the same time shouldnt be a problem no?
I'm not 100% sure, probably depends on the motor driver
but it might be
invalid configuration anyway, doesn't make sense to power both
It shouldn't. At worse, you'll just burn the driver
from what ive seen online, most people power both, directly to the same 5V source as the VCC
like here
ye, my logic would be that the cooked driver could then cause other failures, especially since it has access to the 12/24 V 
in the diagram it's wired up to different GPIO pins, which is where it should be
Hmmm, makes sense
it's a bidirectional driver, so you want to be able to programmatically toggle these pins to control the direction of your motor
and obviously it can't go in both directions at the same time, so having both active at once is dumb














would protect the GPIO at least