#programming
556280 messages · Page 557 of 557 (latest)
i have to reprogram an entire system due to a small oversight :/
usually my errors are missing semicolons or parentheses
this time the error is me
OK chat
What is good things about NixOS?
"reproducibility" or smth
If you break it you can go back to a working generation
Do I get NVIDIA drivers during installer
Noveau by default
Ask one of the resident advertisers
Maybe Gentoo is a better choice?
Banned for spamming
clanker
also probably made by a clanker
-# can i say that here? "clanker"?
I think you can say clanker just don't call neuro that ig
yeah neuro isn't a clanker
whatever made that gif is tho
are we questioning whether it's ok to use clankerphobic language now
That's a lot of clankers
140 clankers sounds like it'd eat a lot more than 5 million tokens
My motto: Block annoying people
If you ask me: Do you care? No, I don't.
What I hear is up to me. Heck you
Your reality is not part of my reality
110 agents 5 million tokens
This is a server about an AI vtuber, I'm worried people would get offended.
There's no way that isn't a bot of some kind
110 clankers??? for just 2 pounds???
1 LOC added 1000000 LOC removed

Claude, rewrite this in rust
i gtg fix my code bye
-# i swear if this guy's still preaching his motto when i get back I'm gonna lose it
Some student sent me code review for luau zombie path finding project
What do you guys think
(Be grateful, it's a student project)
i don't know if that's good, someone else should take a look
Most worst I've seen from code review ppl sent to me was 1000+ lines of code ModuleScript
did they take inspiration from yandev or smth
Calm down, it's student code
You can't compare his code to github demon level 99 programmers like us
nah my code's worse
i can't cook spaghetti but i can make spaghetti code
update to 42
https://www.youtube.com/watch?v=afcFRz81yTg
What is your 42 ?
didn't heir say something about bots always picking green
they ain't a bot but still funny
i don't know what you're talking about
you sure?
is this keno guy seriously larping or smth
anyway i do actually need to rework some of my code, gtg

who is us

Who would've thought unchecked chromium also leaks CPU usage 
Leaks CPU usage as in, it uses CPU for no reason at all (and other reason why my CPU constantly at 80%)
aa
Something about this feels very automated
I mean it's the same thing over and over again it could just as easily be send -> delete previous messages
I saw that happen with my own eyes so

modssssssssss
yo mods is this considered spamming
@rare bridge
screenshots of deleted messages
0/10 because luau
almost 100% a bot
nuked
if it wasnt a bot then well um shit
I'm inclined to believe they aren't a bot given that the timing of their message deletions were human like
sup
but they've been deleting and resending the same message over and over again
so count that as spam and move on with +1 ban count I guess 
#baking ban evasion situation is insane
dead internet theory but it's 12 year olds thinking they're very edgy pretending to be bots
valid point
Luau is peak bro
(not the first time)
luau is ok but i'm a purist, plain lua(jit) is enough for me

Bro is roblox pilled
google doing random side quests again
though their side quests are usually pretty beneficial
isn't another one of alphabet's child companies or whatever doing it
this doesn't seem very google-y
-# meow
your name gets even harder to type day after day xd
why
Because why not?
Some fella even attempt to run steam inside of docker
what the fuck
Bot spotted
⬆️ ➡️ ⬇️ ⬇️ ⬇️
did you spend it already?
Just learn
tested bunch of game ideas so yeah... but according to community there is soon reset to weekly limit as they tend to do that at times
My motto: Trust, but verify (includes myself)
If you ask me: how much do you trust me?
I respond 60 trust 40 verify.
What you heard is not what I said - your 60 is not my 60.
I'm up to me
You are up to you
Your reality is not my reality.
I have this over 400 MB big holocure like game project that has been cooking quite a while with fancy presentation... goal mode just keeps going... also is more organized than me with files and stuff so it reorganized once project started to really get extensive
based vibecoder
characters and enemies are animated... playable characters already have eight directional animation movement... so... of course we don't frontload every damn asset when someone wants to play the game so it even has dynamic loading
Google back then: Don't be evil
Google now: let's release 32 million mosquitoes
excited to read in 10 years how Google's mosquitoes have finally cascaded into complete global ecosystem collapse as i try to purify the acid rain in my nuclear basement
they done released the data center spiders on the sun we're all fucked

what if bacteria mutates 
Precisely, I was trying to work out how much cover it needed to adequately shield it
Is that the hit movie Resident Evil by Oscar winning director Paul WS Anderson
ok bro i guess

Ended up deciding to stick with Gemma 4 31B over 26B because it still seems to perform well with bartowskis Q5_K_S quant and it burns half the reasoning tokens as 26b so it being 60% slower doesn't actually hurt as much as I'd imagined it would
Keeping the MoE around for vision tasks though
im no expert, but the colours they used for their graph is different form the explanation
btw i haven't made a project post for this yet (waiting for next weeks schedule), but i finished my unofficial neuro schedule api 
each lab has their own colour
if anyone wants to use it (or has suggestions), feel free to look at https://github.com/cloudburstwan/neuro-schedule-api
Yes that scene is out of Resident Evil
It's the red queen which was an ai in the movie
fyi this is my first time working in golang, so if the code is dogshit that's why, but don't be discouraged to point out the dogshit. it'll help me learn
I has given it to the sheep and they said "yay useful"
let them know that i accept both issues and prs on the repo, so if they have anything they want added just make an issue!
it's not your fault if the code is dogshit, because that's just golang 👍
Hm
Linus still hasn't reacted to Vedal's server build yet 
might need openHMD
looks alright i guess, you might wanna use your json lib for errors as well
https://github.com/cloudburstwan/neuro-schedule-api/blob/main/src/http/http.go
https://github.com/cloudburstwan/neuro-schedule-api/blob/main/src/http/json.go
and you do a lock around the assignment here but not the file write. if you need to lock there, the file write is going to cause you a lot more issues than the variable so lock around it too
https://github.com/cloudburstwan/neuro-schedule-api/blob/main/src/hub/hub.go#L68-L70
Claude cyber safeguard is pretty silly
Code virtualization? Sure, fine, go on
String hiding? Oh noes better fill out this form (fuck you mean referrers? OpenAI didn’t ask for no referrers)

could you make issues for these? just so i can track them for my own purposes, i tend to forget stuff easily 
for locking, should i write lock only the def and then rlock the file write, or just extend the lock?
actually, with potential EventStream suipport it'd prob be better to lock for the whole function call
extending the lock to the entire call is fine, though i personally prefer locking only around the side effects to minimize locked time, but you need to be more careful that way
also reading the file would need a rlock too, to avoid reading partial writes
we only read on init and nothing else should be writing to that file.. but it is mounted so fair...
i think
nvm it's a docker volume
doesn't particularly matter where it comes from, only the usage. if it's only meant as the initial ingest then that's alright, doesn't need the rlock
yeah it pulls from the file on init, from that point onwards the data is in memory and is written when Broadcast is called
hmm, as i understand it's meant more for mods/etc when someone with perms posts a schedule in the channel, if it's that infrequent then it's not a big deal tbh
Broadcast is called autonomously when a message in #schedule is published, after a parser function parses the message
so approx right
i wonder if you even need that file, on init you could read the last valid schedule message from the channel and work entirely in memory
if your api is down and a schedule is posted in the meantime, you will be out of date until someone posts a new one or updates, due to that file
could fetch like 10 last messages on init with this, and parse them to load whatever is latest right from source https://pkg.go.dev/github.com/bwmarrin/discordgo?utm_source=godoc#Session.ChannelMessages
what games are you gonna play?
Linux is a perfectly cromulent kernel tho 
oh yea also, for networked software use the AGPL license, that requires opensourcing changes for serverside software, which normal GPL doesn't, as it doesn't count as distribution in that license
that is, if you intend it to be consistently opensource and changes fed back
The only differentiating factor is how distro setup their userspace 
good idea

Me when I use busybox
probably beatsaber mostly lol

I used to be decent and can still play alright
there's a few growing pains with alvr at the moment
mostly making it actually work with my gpu
but also there's stuff with launching the app
my quest won't let me launch alvr unless im connected to a network with internet
which is an issue because my setup doesn't bridge the router I have to the wider world
also every now and then I'll probably poke vrchat but I never really got into it as a concept
vrchat is pretty rough nowadays
it depends
i wouldnt join too many random lobbies
but when you join friends its nice
only if you have friends outside of it sure
but the lobbies themselves are mainly full of terminally online weirdos and kids, or people who don't interact at all, or established friend groups who dont talk to others
Pewds just release Odysseus 

By the way I may have gotten a bunch of money from graduating high school so I may be inquiring for GPU pricing soon-ish
752W seems unhealthy
nah thats fine I think
Like over 1000€ much
So yeah I think that can get me a GPU
addicted to gpus
i tried to get the free gpt to do me a shader thing but it's actually stupid so i guess i'm learning shaders now and doing it myself
Wait, you're younger than Sam? 
I don't know

you know about this, i have sent pictures i know you've seen
this shouldnt be new info lmao
Holy moly pewds is making an entire AI SaaS frontend 
oooh, he has a new vid
Use code PEWDIEPIE at the link below to get an exclusive 60% off an annual Incogni plan: https://incogni.com/pewdiepie
⛵ Check out Odysseus: https://pewdiepie-archdaemon.github.io/odysseus/ ⛵
🌎 Get an exclusive 15% discount on Saily data plans! Use code pewdiepie at checkout. Download Saily app or go to https://saily.com/pewdiepie
...
So what is everyone #baking today?
I just decided to make some Matcha. 
matcha sounds great
And it contains EGCG and anti oxidants
i had made pizza yesterday but i already ate it
And caffeine uh
yuh, matcha is just fancy green tea

Maybe if you have cheap GPUs and cheap VR hardware I may consider
This is wild lmao
Now that I suddenly have some money to work with
Purest form of fancy green tea with highest amounts of EGCG
i do wanna go to japan someday and get authentic matcha experience
Oh yea
Fun fact: a chunk of Odysseus was built from a phone -- mobile shells (Termux), the PWA install, and on-device agents. So "works on mobile" isn't an afterthought, it's where a lot of it actually happened.
The dude is cracked 
Look at this nice foam tho
what happens when the average nerd becomes a millionaire
beautiful
I hate the taste but its better than coffee for you
So i still drink it sometimes for the effect
Hm too strong / bitter?
Different kinds of matcha quality usually have different bitterness
The more premium matcha is usually smoother
Might give it a whirl
slap some milk and sugar in there and u got a latte
I mostly drink yorkshire gold these days
Ah no sugar
british alert
Im as American as they come
lies, who else would drink yorkshire
Some friends call me American vedal
Grew up drinking a variety of teas
And after some recommendations i got hooked on it
Finished result
still bagged tho
leafy best
Meh doesnt make a difference for me with this
It blows alot of the others ive had out of the water
interesting
There is probably better out there but i don't really have the free time to try teas these days
that frametime looks bad
is it routed to the right gpu
it feels bad
makes sense, you're getting like 18fps or so

infinite fps 
I just put some Manuka honey in there 
I wish I had my amd desktop to hand
Mh i added some milk as well

SteamOS still exists?
idk whats even happening now with my gpu
I think games are rendering on the igpu but video encode is happening on the dgpu

Public Release of Steam OS 3 is delayed until Satan rises again from hell to earth

I just let Claude code fix my nix configs lol

enjoy walk 
2 full hours of trying to make steamvr, nixos and my quest 3 play nicely

with Nvidia for the fun of it
idk what more I could add to make this any harder on linux
oh yeah
its a laptop 
Use code PEWDIEPIE at the link below to get an exclusive 60% off an annual Incogni plan: https://incogni.com/pewdiepie
⛵ Check out Odysseus: https://pewdiepie-archdaemon.github.io/odysseus/ ⛵
🌎 Get an exclusive 15% discount on Saily data plans! Use code pewdiepie at checkout. Download Saily app or go to https://saily.com/pewdiepie
...
Pewdiepie released an agent harness
Wild
You should check the repo

Fun fact: a chunk of Odysseus was built from a phone -- mobile shells (Termux), the PWA install, and on-device agents. So "works on mobile" isn't an afterthought, it's where a lot of it actually happened.
He did use GrapheneOS
is this huge or nothingburger
Lmao
It is neither
Wasn't on my bingo card
To be fair, its likely does not affect him
i hope it isnt his first time on github :V
Surprisingly considering its developed by Pewdiepie of all people it does have a better memory system than a lot of vibeslop harnesses
I highly doubt he's exposing it outside of his LAN so its not a huge deal
Obviously not good but not catastrophic so long as its insulated against CORS fuckery
He likely uses Wireguard anyway
I remember his earlier vid of his journey. Bro really coded up his own spin off of photoshop than learning GIMP
there's exactly 1 thing exposed locally and it's the searx engine lul
the rest will bind to 0.0.0.0 which is wan
Not necessarily port forwarded
Claude will pwn back
if he's hosting it on some more powerful pc to access it from his other devices, it likely would be
Also, 0.0.0.0 is not WAN. It is just every interface connected
Nope. He has 4-8 gpu
Running on lan via wireguard
When he needs it remotely
i wonder how well those "delete my data" services actually work. cuz at the end of the day they can just say the deleted it and still keep it. and the company that asks to delete your data now also has your data
who knows what vibecoded crap he's got set up ¯_(ツ)_/¯
It's the same as how you hire a lawyer to do your legal stuff
they collect your data and sell it for a higher price because now the data is real
They will give you the report

i would like to selfhost more stuff, but most of the services i use cant be selfhosted
Such as?
do it like pewds, vibecode it
streaming services, music app, ...
you know like the daily stuff
You can actually self host those 
Just with your own owned media
i could host my own github locally and ai locally, and thats basically it.
streaming is basically not possible to do locally without downloading the content sublegally
or buying the records legally
Ye this
so stop listening to every song ever made
To be fair, most modern artists didn't launch their album physically anymore
the whole point of streaming for me is that its easy to listen to a new song or watch a new movie
i cant spend hundreds to get the bluray box set of every anime i watch
Its only WAN if you make it so at the router level
Yeah, the value proposition is very dependent on the user 
wouldnt be the first time
For me, I don't watch as much movie nor anime
But I do watch youtube
So I pay premium
even hosting ai locally is pretty bad with how limited the context window is, i can barely send 5 messages in a row before the 8Gb is full
tbf gemma 4 yaps a fuckton
Fun fact, LTT did their breakdown and find that premium user, little as they are, actually contributes more to the revenue than ad supported viewer
Peak?
That means paying premium actually do helps your preferred content creator
THE ONE PIECE IS REAL
02 ❤️
i pay for yt music premium. and i just dont watch ads on main youtube
i ocnsider 6 euro per month to be enough for them
Use 26BA4B and offload some layers then you'll be able to fit plenty
You can get like 30t/s even on CPU with it
You get music premium bundled with youtube premium
im not running pi hole actually. but i could i guess.
more expensive tho
xpotato
That's fair, but if you are paying family plan then it is cheaper
what the fuck is this?
He got a goal
im not family sized
im sam sized

;)
I mean you do have a sister don't you?
i do
Split the cost with her 
the promised codex weekly use limit got reset... let's go! https://puu.sh/KNOBR/903a16a80d.png
Or run a smaller model like Ministral 14B
the promised land is real
PAAAAAAAAA
Or even Gemma 4 E4B if you're not relying on internal knowledge
im running 26B A4B it heretic 4bit
on 4090
Why is there no evilpat 
you nurpat2
Omw
Then something's wrong, I get 50t/s at Q8 with this setup on my 4090/7700X system
mmproj = C:\Users\pikac\AppData\Local\llama.cpp\mmproj-google_gemma-4-26B-A4B-it-bf16.gguf
ctx-size = 32768
ctx-checkpoints = 24
n-cpu-moe = 8
fit-target = 512
temp = 1.0
min-p = 0.00
repeat-penalty = 1.00
presence-penalty = 0.00
repeat-last-n = 0
top-p = 0.95
top-k = 64
flash-attn = on
mlock = on
no-mmap = on
reasoning-budget = -1
parallel = 1```
neuropath2
where the hell did you get that list from?
I made it myself
Chad
Its in my llama.cpp models.ini file
Honestly you'd be better off using llama.cpp
these are named so badly
LLVM is the compiler guy
everything has the same letters
Way more flexible, VLLM is better only if you're relying on tensor parallelism or need to optimise for a multi user setup imo
Yeah but I doubt you serve all 10 at single point in time 
Use my philosophy of crash early 
If you're not batching then llama.cpp is no better or worse in the first place tbh
Maybe a little less raw speed but you trade that for stability, much more sensible config options and way nicer quant options
I really wouldn't rely on a 4 bit quant of a 4b active params model
Wait why are you not using Q8 and a longer context window
can confirm 
it is dentge
at 4-bit
Even in VLLM that should be quite do-able on a setup like this
i have no clue what that means
ask your llm to fix itself 

The model weight quantization
And the context the model has access to
what about it?
Q8 = 8 bit quant
Potat is saying you can use Q8 instead of Q4 with your setup
Why not offload with a model that has so few active params?
I recommended llama.cpp but you can do it with regex in VLLM too
offload to where?
no
hmmm
i odnt think they let me put layers on the cpu like someone mentioned to me before
If you really insist on fitting the model entirely on GPU use a 4 bit quant of the 31B model with 32k context at least
It makes no sense to use the MoE if you're not offloading
i dont really mind whether its on the cpu or gpu, but fully cpu is probably slow as fuck
its pcie 3.0 x16 so not the fastest
Oh okay yeah you probs don't wanna offload
Not for Q8 
In that case grab this version of the 31B model and use 32k context
It'll work much better within your constraints
kniipedia, what mean?
that almost reads like a hex number
covfefe
The 31b also suffers at such a low quant but it'll be way, way better than the MoE if you need to keep the entire model on your 4090
its an investiture with power, estate, stipend, or title within a feudal government

i vote bullshit on that word
unenfeoffed
holy shit thats a real world
hardware is 2x xeon 6138, 256gb of 2133mhz ram, and a 4090 on pcie 3.0 x16.
i want it to be able to swear, and type at least somewhat fast
second word of the day
investiture
fiefdoms
vision would also be cool
train an ai to say only "fuck" has only one neuron saying "fuck"

its a normal gemma 4, but it uses MoE to swear 
lu
"which slur in this context would max out the offensive the most and be closest to the eigenslur for this use case"
;
we inverted an llm to find the most offensive words. the results will send you straight to literal hell. there is no way back, we tried.
once you can swear you can slur
not really
so youre speaking from hell
yes
and once you can slur you can definetely swear
The base model can already swear
Abliteration also nukes model performance
the words we found offended god himself. also yea we discovered that god exists too
i was hoping for the copypasta
its very hesitant about it, but it is capable of doing so it seems. its using the not so bad ones.
i also tried with swearing at the model instead of getting it to swear and it seems light swears work and speed up performance but move to cursing bloodlines and slurs performance drops and it starts worrying more about pleasing the user and maintaining peace than working
Unless you're roleplaying with the model (which isn't something I really want to aid in, gives me the ick) the base model is absolutely fine and not at all overly censored
such a gentleman (ive seen some horrific shit said from chatgpt 4 and 3.5 bots on the internet)
i don't think im allowed to show the evidence. it only used the word for the mentally disabled
IDK why people like roleplaying with soulless thoughtless machine
bro i cant roleplay with a real person over chat roleplaying with a clanker is boring
What models do the role play places use anyway?

Why would youe ven want your model to say slurs
It might be funny once in an edgy 14 year old kind of way but there's no practical reason to make that a part of your daily driver model choice

no i swear at deepseek in cantonese to get it to speed up
occasional say slurs but that seems counterproductive to work


wait, you're cantonese?
Maybe be nice to the soulless machine instead?
I mean you don’t swear at your computer right
(maybe some people do)
Cantonese is not really a natioality 
its a region no?
it’s a language
It's a language used from a certain region 
canton = region
Or maybe dialect I guess
That's canto 
jester is ash ketchum

I will say, you can see some quant degradation here 
Okay hold up
Maybe if I tell to recite the original it'll do it without error
Okay yeah this quant is showing its limitations
Time to go one tier up
Wait no bf16 ends up reciting variations on it too tbf
That stray character is disconcerting though
sam



ur mom
what about my mom?
not yours


im adopted?



now this is just cute lol https://puu.sh/KNOGF/1f90c613a0.png
I had a realization... I can have better website than my current mess...
so that's being cooked on lol
your realization is you can improve your bad code?
my realization is that I should replace my website builder made site with better one
wouldnt this use 18GB of vram and then id have less context window?
virtual dom maxing
small typo 
I can fit Q4_K_L at 32k F16 context
This is actually smaller than that
What I'm saying is you can fit at least 32k tokens worth of context on a 4090 with that model
codex feels smarter after global weekly reset but it might just be placebo effect
virtual DOM maxing?
tung tung tung deez nuts
Tripple T
wuh does DOM have a different meaning

i plead the 5th

i plead 5 buckshots into everyone's heads here
I would take the fifth
Probably placebo
you just became dumber
Honest truth
If u start relying too much
Ur not forced to think anymore so u dont
Or maybe they serve worse models depending on load, who knows
…how feasible is doing that?
as someone who overthinks at times... I'm still actively figuring out things regardless
5 liters of water saved 
Still a way to reduce the gray matter in ur head
I will drink them myself
my gray matter is busy learning various things... I agree when it comes to people that want to delegate actual thinking to systems without oversight that are worrying... like they're still figuring out how to get through schooling
lotta people schizo that they serve worse and worse quants depending
there are also 3rd party providers that pretty much only serve quants
Honestly wouldnt be surprised
Open-source evaluation runtime for testing CLI-based coding agents. Measure accuracy, tokens, duration, and capture full execution traces.
New gimick dropped
New technical difficulty dropped
sloptuber
Good morning
first post of it

yeah i just know it
I'm not the only one having crackling audio output right
still confused how to instal it
vedal issues

DO NOT put the microphone in rice
It sounds like the audio is skipping frames or something
Bitrate mismatch?
is this what the big audio want us to think
why is my Twitch dying
how did they get the drums in there
Could be the CPU
Well that's a secret
New cpu is too fast for the audio encoder so it skips hz
I decrypt it
What's the password?
the return of the turbo button

Remember that time where suddenly games ran at 300% speed due to too fast clock speeds lmao
Had to increase my delay a bunch before on my ASIO
When mixing some big projects
ASIO? wasn't that a windows thing ?
Its just a specialized audio driver
I don't hear crackling
I wonder what he changed
Yeah do you use Windows or Mac big projects lag so much I swear 😭

No but now there are pauses
partially fixed mhm
raaaaah
See that's why Vedal need NixOS
If he used nix he had a frozen working state in his git tree and could revert any time
I do not enjoy linux shenanegans
And I’m here too
I tried to fix my issues with an llm, claude is losing his shit
Honestly I had Linux mint, but playing games is a nightmare so I had to switch back 😂
Claude could be a girl
Hm usually nowadays game compatibility is quite good
Are you having issues with a certain title? Online games with unsupported anti cheat are still causing issues.
Claude could be the exception
But yeah no your right
i just wanna play beatsaber 
Thats literally worse
Yeah exactly most DBD, marvel rivals and overwatch terrible for launcher support game would crash on boot up or take 10 years to load shaders. Tried to fix it, but every time they updated the game I'd have to wait a few weeks minimum.
If I ever have a daughter I’m naming her Claude to spite you
For that issue
I am never going back to arch
U have alot more bs u have to do
I prefer perfectly documented BS than undocumented BS 
We should all use Hannah Montana Linux actaully
I assume you already checked ProtonDB, AreWeAntiCheatYet and Github issues of Proton ?
i never watched that
It aint worth the effort for ur average employed guy
As a Java programmer bs is my job
Me when developing embedded system on windows: 
thats on you for doing java.
cant say much as a gamedev myself tho
Do you wear glasses?
Plenty of other distros
Yes, but you'd usually have to wait for driver updates and that can take like 3 months depending on compatibility with Marvel Rivals I genuinely have nightmares from using Linux now 😭😭
No, but I probably should
Yes, but how many of those has good documentation? 
I thought because you can't C#
Documentation isnt always needed if its built intuitive
Which arch isnt by nature
"Built intuitive" 
Haha no I've got C++ extra reach on that 😎
and crackling is back
This simple mfker has 2 interpretation
Dated
I swear they where like 100 bucks a little bit ago
they were yep
The prices are only continuing to go up
People found out you can get 24 gb vram for OpenClaw for cheap

Seems pretty obvious to me
Fucking OpenClaw
OpenFlaw
wait aren't tesla cards like old
we cant catch a break
The waifus are worth it
Yes like over a decade old for some
Can we all agree computers are scary and ahhh
why the hell is a card from 2016 $350+ now
But they used to be cheap and decent enough for a budget home system
yeesh
your "AI waifu" is just selling your data to every company on planet earth to sell you skincare product
Did u miss the AI requirements increasing chip prices?
No, I knew about that. But these where so old that they where 100 bucks for years.
but imagine how glorious it will be when the ai bubble collapses the economy and you will be able to buy 20 gpus for a liter of water
which I guess must be steaming over the network?
If it has a memory chip its going up in price
Haha she would do that
What a little prankster
actually, im in full support of ai waifus if it makes more people do self-hosting
Worst time to self host rn sadly
But i do encourage it
Especially as everything is all cloud now
Which is horrible
It's faster cousin with half the ram is now the price it used to be
Have you read arch wiki?
i already have the hardware for running decent ai models, what could go wrong?
I have 8GB of ram

So idk if I can host shit
You need more 😂😂😭😭
If it aint even deepseek levels then its not really even worth it
I do. Now gimme your hardware. This is an online mogging 
Guys is 8GB of ram enough to run a sentient ai
Just a really expensive markov chain
every computer on the planet isn't enough for sentient ai
good point I forget the arch wiki exists
But it has 8GB
Fml
you can have 8 trillion petabytes of ram, doesnt matter.
no sentient ai exists

Feed my caffeine addiction 
I wonder if its worth just grabbing two P100s, thats 200 bucks before shipping and 32gb hbm2
I think 8GB will be enough 
Why is Neuro now in Evils room 
And I don't think it ever will
also, thats ram. how much vram you got?
2
thats like
But you can mog on people with ur 8 trillion petabytes of ram
😭 ✌️
the worst arch wiki page ive ever seen
The best people can do is smoke and mirrors it.
mogging on people with ram you're not using is just sad
The issue on-stream is probably a Windows update
Fml
one day i will create an agi, yall will see
That does not explain why Neuro is suddenly in Evils room tho 
wouldn't be suprising
windows corrupted the files
in the update
Speaks volume to how supported VR is on linux 
and failed

No, but you might pull a real girl
That's on Ved, but the audio is definitely on Windows
6GB I think
I still don't understand why he does not already has switched to Linux
with the rewrite windows is like alot less necosary
girls dont care about unused ram. unused ram is peak impotence
cuz he wants to able to play games
is neuro running on the server now?
I thought my sound system was breaking
don't think so yet
still on old PC
There are also good games that work on Linux
Because Ved stoopid and hasn't seen how obviously required it is
this tbf
Lol
Guys is 6GB or vram good enough
I always pushed for native linux releases when i was working as a game dev
Depends on what u need it for
(No mac tho fuck them)
valid
good enough for what
Almost everything runs on proton these days except some MMOs or other anti cheat games
W
Running my ai child
But it's omega and girls love to omega on these nuts
and neuro cant run under AC games anyways, because of reasons. And VRC is an exception which they are already allowed for anyways 💀
then no, probably not 
Gaming? If you go a few years old or don't play UE5 games
ML? Absolutely not really
So there is literally no reason to stay on windows
yep
If u play modern UE slop then good luck
VRC also does work natively and just runs AC in user mode
so
even if that was a thing
U can run like sum 6b models I think idk much Abt ai
dosent matter
WE are not getting back into the AC debate today lmfao
there is a reason
compat layer makes everything more complicated 
we are making fun of ved for being on windows
That should work via Proton as well
believe me, as someone with 320GB of ram, girls really dont care
VRC has a native Linux build from wat I know
being technically feasible does not automatically make it a good idea
TF U MEAN 320GIGA BYTES OF RAM
WHAAA
Just change today's stream to a "Switching to Linux" surprise stream instead
Lol even better
DAS LIKE BILLIONS
ong
tbf he can do software atleast semi competnetly
Atleas
tsemi
He managed ot make neuro
it was like 100 euro
At least he's better with software
Lmao any mod pls do that
ong
Opens every port to public
😭
Switching to Linux now required for Neuro streams
Gotta port forward telnet
linux
with recent windows update yes tbh
Best stream ever
Something with a recent Windows update must have messed up something critical
#programming try not to blame everything on Windows even when the issue is completely unknown (impossible challenge)
windows is nearly always the random issue
statisitically
I cant spell
yeah
no lol
How much percent do we bet it's windows fault ?
I want a gamba pool
tnow
101%
the reason linux fixes the issues is cuz you're forced to use a different app on linux
cuz nothing supports linux
Microsoft litrally fucked up notepad and had an rce in it so YES I WILL BLAME WINDOWS FOR EVEYTHING
Not really 
im exaggerating, but the motherfuckers at adobe grrrr
Lmao i love RCEs
As long as Ved keeps for whatever reason using Winslop we will continue to blame major unsolvable issues like this to him using Windows
Yeah, mostly adobe. Funnily enough, Fusion360 works 
On Tinder I said I have 160GB of ram and she sent me a winkie emoji
But also, Affinity Studio\
tbf if Neuro SDK just works by opening a port / web socket connection with the SDK or watevr, [I am not familiar at all with the SDK], it shouldn't be that hard to work under proton either way
source?

calling that one an "RCE" is certainly a choice 
it allowed links with protocols other than HTTP, that was the whole issue
I would sell mah soul for sum 8gig ddr5
Should be basically plug and play no?
you're asking for too much. the real value is at ddr4 ecc
still crunchy
SSRF?
file:///etc/passwd and funny gopher:// chains have entered the chat
well
bru what
Alright, this is probably not windows issue 
Neuro has sore throat
issue
ong
EVIL STREAM 
Same it's built like gundam makes my pants go flying
Lol my stream was stuck on twitch offline screen
EVIL STREAM 
I will still say it was windows, windows just corrupted neuro trust ong. totally, windows causes all the issues trust me bro
vedal gonna write his own OS for Neuro
this aint even on server yet
if a message he sent the other day is to believed
I do wonder what is the cause if Evil isn't scuffed
Nah I'd win. Dms are open cash in hand or sloppy toppy
yeah pretty much, except the server is notepad
so you could access any protocols that some random app registered, like the steam:// one registered by Steam
so I guess you could open the calculator or make Steam download a game when you click on a link 
incredible RCE
neuro's got that nervous tick of someone about to be transferred to a new server
wait
Wait what, was the Ebil run a test to see if a fix worked?
Funny
No, most likely to test if Evil is also broken
keep it at least somewhat PG13
maybe a sanity check tbh
??
It's still borked
Poor neuro tho
yep
crunch is gone but she is stuttering
he fixed it?
what did this come from btw

n




