#programming
1 messages ยท Page 547 of 1
honestly, he should have died, granting him a free pass just made him a permanent mary sue
the regen has no point anymore, he is by all means actually invincible and immortal now
But then the show would have ended
it would have been a good ending
they could have started another show in the same universe like torchwood
theres an endless amount of incredible characters there
like you still have master, you have his wife, etc
There already are the Sarah Jane adventures and torchwood and class
So a lot of spinoffs already
yea and theres plenty more they could make, its a treasure trove
Sure but nothing is like the original
and like, you can still have the doctor tho, because its timetravel
Ok and then Chinballs ruined the show
but he shouldnt be the focus anymore, its too much
i miss the selfish doctor who would commit genocide for those he loved ๐
I was hoping RTD could bring that back
gpt models love randomly dropping implementation details into the layout. like bruh what why put that there what
this is a working better auth
.. shut the fuck up
they are prolly pressured from up top to be "politically correct" and safe now, driven by various interest groups
capaldi era pretty much started cracking hard
lol what do you expect from a clanker
The political correctness just destroys everything
to not do that which shouldn't be too hard considering gpt models are unique in this
they're all gpt models
openai models
meh, i've never seen any of them do anything coherently
Everytime I try gpt again Claude opus is miles ahead in about anything lol
and i am miles ahead of any of them

idk how they keep managing to make bun worse
"small change, ez review"

and i am miles behind any of them
from what I've read it doesn't even really help much
soooo they're moving away from the rust rewrite they just did a couple weeks ago???
so now you need 3 different compilers to build bun right
it already had CPP stuff apparently because of v8
man i detest the js ecosystem
just don't use js 
forced to if i want to make a proper website
Its not nice to lie
fake news sold by big JavaScript 
i'll never let js on backend though
What even is Bun? ๐
js runtime that prides itself on being fast
bnuy
Ah ok
if you need dynamic functionality for web based software, ya you do, or the interactivity is complete ass
i only write vanillajs though
Bun is as good as dead
got acquired by anthropic and they vibe translated it from zig to rust because zig banned ai contributions or something
Lol
Based zig
it is not a lie
Ive been migrating my stuff back to node
I'm also just vibe coding 90% but that's just because i can't code
smelly vibecorders 
it had a great pr
not codder
Vibe typers would be a better name
won't even load the image preview 
100% not programmer
Rebrand vibecoders to vibetypers
Eh i can do other things good
rebrand them to vibrators because they move in circles and disturb others
But yeah its a shame we lost bun
There is another option i cant remember the name of besides node
guys, should this fail, or my code is wrong?
isnt it should just swap buffer back and forth?
idk really, its all none, but here i expect some
Starts with a d
what's even the goal here? what's the failure?
Tbh i still do use it for discord bots
Hey but prototype pollution and pug/ejs SSTI is fun
But thats mostly because its pretty locked down in that area
I do wish we had a decent cross platform language for objects for web dev
does djs even have docs
highlighted assert
buffer is mutable...
i expected that it will result some packet, not none
since buffer should be in state that it was begin with, since buffer contain both nonce, auth tag, and cipher text
Yeah
I looked a while back and only found basic docs that told me what args methods took
Might of changed idk
but maybe gcm buffer somehow disallow using on same buffer, from internal state of gcm object
I wont ever use python for production
but you'll use js 
For web related stuff yeah
Using Python for production is like using matlab for production
calling discord bots web related feels like a stretch
if i understand the code right, you're creating an encrypted buffer raw and then modifying its raw encrypted contents afterwards? that will never decrypt back out so i would expect none
It essentially just exists for the API
Not really anything desktop related
why iy will never decrypt
what gcm do on top of aes + auth tag
isnt it supposed to have no state? (except initial key)
ig but like
Plus its just being looked at by discord itself
to me web is the Dom and stuff
If it was some public backend then yeah
But otherwise it doesnt matter too much to me
actually why are you handing a slice [4..] to the decryptor? wouldn't you want the whole packet data?
i don't understand what you're even trying to do there, your api or its usage doesn't make sense
๐
first 4 bytes is whole tcp frame data, it is stream
i decrypt on array with size from this 4 plaintext size
padding to aes block inside buffer; with some random
i would expect the api to work symmetrically, if i create a Packet::into_encrypted i would pass the exact same result to Packet::from_encrypted and get out what i put in
i would never expect it to work if i modify the encrypted buffer/data or pass a random slice
Stylised graphics on a modest polygon budget were a special thing
Modern titles aping the era always feel the need to cheat
Black mesa vibes
that apparently is it
but packet function not modify content, decrypt function use in_place_detached
so i not understand what extra happens here, i assumed it will just use same nonce -> produce same result not matter how many times i applied decrypt twice on buffer
nonce will be part of the encrypted result
Reminds me of when i built my own mc server
so nonce shouldnt be the issue unless you screwed it
Gaming netcode my beloathed
if you modify the encrypted contents, i would not expect it to work well in GCM either, as the chunks depend on eachother for correct decode, so i would expect a failure/no return
every time I hear MC server netcode get mentioned I am obligated to post the macro I wrote
Lil different from how i did it
Mostly because i needed java parity for certain things
Which meant i had to recreate Java bs in C++

the ones that annoyed me the most were the "just take the rest of the data in the packet"
because I couldn't find a good way to write that in a way that's impossible to fuck up
I could shit on minecrafts horrible netcode all day
Thankfully its not as horrible as it once was
gah i see the first an issue
i in result modified encrypted data to 0, should've save initial state
but that apparently still cause assert at same line
not it cause assert on second some omg why
now it passes first assert, it decrypt to valid data once
but when i modify buffer again, it fail
i dont understand what happening
The amount of horrible desyncs and packet duplications before though were enough to make ur blood boil
In cryptography, Galois/Counter Mode (GCM) is a mode of operation for symmetric-key cryptographic block ciphers that provides both encryption and message authentication in a single pass. It belongs to the class of authenticated encryption with associated data (AEAD) algorithms. In practice, GCM gives a recipient two guarantees at once: the messa...
because gcm
do not modify the encrypted data, it will not decrypt
Damn, using this game to test out lossless scaling frame gen and it has come a long way
GCM encrypts in a self-validating chain of blocks, if any data is modified the whole thing fails
read again
i said already first is_some pass
and when i comment second is some - whole test pass
but now - why second is some fail?
Why are we despair reacting
Frame gen is a genuinely nice utility for games with hard 60fps caps on OLED
also that kinda whole point of testing on encrypted data
to see if anything corrupt if you mess it up
apparently all (manually fixed tho) ~37 ai generated tests pass
now i want to write some manual, just in case...
the more i write rust the more i actually like the language... atleast compared to the alternatives
cargo is a mess tho. why do i need 50 deps to make a simple app
where's this api from? sounds like it decrypts in place?
because it is
if it does, then it would fail the second time, because it discarded your encrypted source
Like yeah it does increase input latency but its so low on an OLED that it hardly matters in a singleplayer title
The vsync era was way worse and didn't come with the added benefit of hugely improved motion clarity
so the failure is expected
no one discarded it
no, first assert pass
only mutable op here is rust crypto aes gcm impl call to decrypt in place detached
print out the contents of the decoded buffers
i guess so this is the best way rn
counting 
if it decrypts in place, first assert will succeed, second fail as now your buffer is decoded and no longer valid ciphertext
and third should success too
but at second decode you made it ciphertext again
if you reencrypted it in the same buffer, sure, it would work
but you dont, so it fails, as expected
print the buffer out between/after each call
you think it count auth tag first, and then mutate buffer? or partially mutate it?
both not sound reasonable
no idea, never seen that api
implicit in-place decrypt sounds insane i don't know who would write it that way
in place decode?
me neither, but that seems reasonable to decode on buffer, and just return whatever auth tag valid or not
BUT
it work when i decode and fail
and then decode with valid ciphertext
even when i decoded only once
tried*
yes because it modified your source buffer
the source buffer no longer contains a valid decryptable message if it succeeds, because of that api
so, apparently this api modify buffer after decode, or something like this
i not see why it is it
since owning api (that return to you some vector), create vector, and use inplace method on it
i avoided allocation anyway, but i not understand api
also
encrypt_with_backend
it is template hell
i not found implementation without docs
??? i don't understand what you're saying, did you write that api?
idk how it even looks 
no, it is rust crypto aes
that's not what i'm talking about, who wrote the packet api, that's not part of the crypto package
so, this thing work
not %2 amount of failed decrypt calls
packet api is mine, and i call only in_place that is mutable
no fuck
ai offered me using 12byte nonce for ctr aes
okay, so you understand that inplace decrypt will destroy your encrypted source data
after a successful decryption
GPU is trash, but I guess good enough for 200โฌ
that's a nice system for 200
I guess is okay machine for running bunch of low power crap too
Can even put a 5800X or summat in there
-# nope,
why it should destroy data anyhow? supposedly api has only provided buffer, it is normal to assume that cipher will be applied to it, and returned only whatever auth tag valid or not
do you understand what in place means?
it means literally in place; why would it use additional buffer if it can just not
do you also understand what mutable means
are you joking rn?
you are passing it the encrypted data buffer as mutable, in_place modifies the buffer, your data is overwritten by plaintext
doesn't matter what you return, you modified the buffer you passed in with in_place
there nothing about it in comments
about additional buffer or special behavior on failure or whatever
๐
have to read docs/source
but when i was without internet, i not found encrypt_with_backend anywhere
so probably wouldnt find decrypt either
it is literally not match anything in lsp find references
fuck it, just add .clone() to the buffer you pass into your from_encrypted
there i fixed it for you
but it seems like it is not overrides, when auth tag is invalid
that means api use additional buffer
why
what is it even supposed to do here lol
there no fix, nothing broken, just api work in weird way
clones your buffer so you dont destroy the original on a successful decrypt
it doesnt work in a weird way, you just don't understand what it does
except that i overrided ciphertext with 0 instead of saving state initially
api not break buffer on decrypt
when it fail
yes because it validates first and then decrypts if it's valid
apparently buffer just not changed at all, when it fail
how it validate? isnt auth tag calculated on whole data during entire decryption process?
aes is just xor
there's so much wrong with your assumptions about how the cryptography you are using works, that i can't really help you understand right now
either read on what aes and gcm does and the crypto api you're using, or understand that your usage will replace the source buffer with plaintext on success, so work with it accordingly
integrity by mac calculating on not decrypted data is against what i understand for sure
i implemented custom mac once (for custom not for real use cipher), and i used entire decryption process to get result, would be surprise if it can be done other way
poor guy
It feels like this started happening more frequently
https://www.phoronix.com/news/ANGLE-Merges-Wayland
I'm a bit confused, I was under the impression they already supported Wayland
man was so active
Someone abuses mythos to break swarm accounts 
it comes and goes in waves
people just want some free cash
@silent cloak I changed my mind its pretty bad once you have a gun to occlude things
lossless scaling is great for older single player controller games 

wrrr

spent a bit writing some code
only to remember that once again I have been foiled by my own laziness
I really need to finishs moving my entire pc to flakes
(currently blocked by rewriting my nix config, which is so good for procrastinating)
Where ?
Oh what a shame
Also if your on a limited budget getting extra frames by getting a older GPU to offload the frame gen to

AI bros of #programming, if I had a list of thousands of discord messages that I've sent, how would I go about fine tuning a model so that it speaks like me, all the tutorials I've found is for cuda and I'm on an rx 6800xt. The model will be used for an afk bot that will respond on my behalf so it doesn't need to mimic my behaviour, just the way that I speak.
Also if you have a model you recommend that would be great.
I remember hearing about someone getting x4 framegen ages ago, because they had a 40 series gpu doing 1 set of framegen, but then had an amd card doing fsr framegen on that output and then dlss framegen on that frame again.
You can do it
Input latency go brrrrrrrrr
It looks like a DMT drug trip
How would you know what that looks like?
https://www.youtube.com/watch?v=GDvfIbRIb3U My favorite doomposting channel did a good video on it
Sponsor: Montech HS01 Pro on Amazon https://geni.us/UYLeZn and HS02 Pro https://geni.us/ABM9fEB
Today, we're benchmarking the $7 Lossless Scaling Frame Generation tool's impact on total end-to-end system latency when generating frames. Part of our latency benchmarks today include dual-GPU configurations, using one older GPU to offload some of th...
Join us and over 70 million players in one of the most comprehensive vehicle combat games EVER MADE! Check out War Thunder for free today on PC, Console, and Mobile using our links!
PC/Console: https://playwt.link/linustechtips
Mobile: https://wtm.game/linustechtips
Check out Lossless Scaling: https://store.steampowered.com/app/993090/Lossles...
Oh it probably was a legitimate account
Something like this happened even to me some time ago
I just accepted a friend request on discord and that was enough to get victim of a token grabber. Next thing that happened is that a similar spam message got sent to every server and every channel i was in. And i did not even do anything.
Chances are high that this guy is actually innocent.
Can't help with the part about running the training on AMD, but in regards fine tuning one of the simplest methods would be getting pairs of inputs and completions for the model to generate, that way the model gets the best context on how to respond to stuff like you. But if you want to train only on your data alone, then I guess you'd have to have examples be entirely masked for completion in training and you'd use a custom/modified trainer (idk if there are any that doesn't assume default shape of turn taking with input and model output).
I have had the dumb as shit idea of buying old 10 seres and rx 4/500 gpus ot see what kinda performance you cna get with like
Dual GT 1030 GDDR5s
Obviously a render, the animation is good but too organised in my opinion 5/10 
Dual GPU does not worth the hassle 
MAC is just a way for you to authenticate a message. So kinda like making sure you know the data comes from the source you know 
run it through sha256 and you have your (h)mac and done 
this is a lie actually, hashing itself is not enough
Nope. MAC implies you have a shared secret 
fine, then xor it once after hashing

hmm.
It looks like those hacked accounts might have something to do with cheats?
i forgot the couple properties a secure mac system has to hold 
It happens. Crypto system is... a lot to take in sometimes 
I toasted cryptography, and ate it
it was yummy
what are you all talking about

โผ๏ธ

Jinja2 templating is oftentimes dangerous if not sanitized properly 
Every templating is 
Also, man I wish OCB were the common mode of operation instead of GCM 
and this is why it's important to sanitize any user inputs flowing into templating engines lol
RCE is a piece of cake from here on
Three Buddy Problem x Ekoparty Miami Jordan Wiens on AI, Offense vs. Defense, and the Dying CTF Pipeline YouTube Spotify (with video) Apple Podcasts Find your podcast platform RSS feed SHOW SOME LOVE TO OUR SPONSOR TLPBLACK provides frontline defenders with high-fidelity threat intellig...
JAGS (04:51): I'm so glad that didn't happen. Ghidra is a wonderful thing for the people who didn't have access โ I'm happy for them. But I don't ever want to fucking use Ghidra. I would like to spend the rest of my life without ever having to use Ghidra. Java โ I don't want to use that shit. I don't want to figure out the 17 different screens I need to click on, because that's how it was built up by a bunch of contractors over two decades. It's wonderful โ but I'd like to never have to use it, please.
Jordan (05:33): They dropped two clicks โ it used to be 13 from literally launching it to looking at main. I counted, and it's down to like 11 now because the file-open got a little better. It's a product built for long-term reverse engineering: you're inside the NSA working on a target for years, decades of reverse engineering. For 'I just want to look at a thing quickly,' it's not what it was built for. I'm glad it exists in many ways โ it's good for the community. Radare was the previous tool of choice, the only open-source tool, really.
Audible chuckle

god fucking damn it...
can't find a way to install the library either, doing this via remote terminal
note to self: do not run git filter-repo when you have unstaged changes
nothing important lost thankfully 
i once lost a whole repo with my filter git nonesense...
2 weeks before the deadline..
i was trying to filter out git lfs files from my git history so i can back up the repo to github... i didn't have enough disk space to duplicate the repo.. so i just did it raw on the original only version of the repo i had 

i exported the project a few days before and was able to recover the assets that way. the code i had to recover using vscode's timeline feature
did you guys know it keeps track of every change you ever do? even if you delete the file? the files still exist
git filter repo good to fake that you did work earlier or later than you did
I have git tree
but that's the git cli for you i guess
i just use lazygit nowadays
But way too much
i would use gitbutler for everything but they're still working on support for a workflow without branches, any day now 
I wish git would have a better submodule worflow so monorepo is not a thing 
submodules really are just eh... i don't like working with them either
It is worth it kinda. If you have a low or mid end system, then getting a newer used gpu, and using your curren one as the frame renderer. You got 2-3x the frames and minimal latency increases. It just a bitch to setup sometimes.
Exactly. If their ergonomics were better we wouldn't need package management 
but aren't package managers just the better ergonomics in the end?
And look at how many security issues it has caused
i mean... all rust is really doing is just pulling the source code and then building it with your project. just like a git submodule would
the same would have happened if they were on github and we all just used submodules
Yeah but it is too opaque if you need to vendor a patch
Hasn't gitbutler succombed to the we wanna make a product for AI and not for humans anymore ?
They fucking sucks so bad
it is indeed a bit opaque. but the technicallity is basically the same 
Yes, now you should try to apply a patch first before building using package manager 
the worm would just publish new git versions and people would pull those submodules intead of cargo or npm packages
i just fork the project. since 90% of rust things are open source anyway on github
I doubt signed commit can be replicated easily by just authentication
and then we are back to submodule 
Too good ergonomics make it back
Imo Java is just at the right place
It's hard enough that it's annoying and you consider things
But it's easy enough that it's doable and you don't need to just vendor stuff
I never have a dependency explosion in Java in fact it's quite the opposite
I'll just have the least i can while reaching for things that kinda really need one
That's exactly the issue. Why should you need to fork a project to make a one time patch 
build.rs ๐คฆ
No install scripts we said
because even just adding the patch is technically a fork
You should PR it instead
That's true. But then vendoring it is harder to track 
build.rs is the most stupidest thing ever
why does it even exist... or rather why is it enabled by default.. so dumb
The patch may not align with the upstream view
Then make your own project or maintain a downstream
I fell for a phishing facebook account grabber email 

Not worth it. One of the OPAQUE authentication library only needs a simple patch where they don't embed the fucking WASM build as base64
would be news to me, they haven't really added much AI related stuff beyond the commit msg generation and their weird claude code integration that exists for some reason
their whole website is all about AI 
Tbh maintaing downstream with the right tools is easy
Just do not for the love of god use merge
Its a burner account they won't get anyhting useful. What will they get? Some anime memes and my marketplace conversations?
Merge when maintaining a soft-fork is evil
It is easy BUT as I said, it is not worth it for one off patch like I mentioned
A soft-fork should only ever use rebase and git rerere
Or when I change the crypto suite of upstream
Is it tho
You don't need a patch for that unless downstream is badly done
is it? the parallel branches stuff lends itself quite well to agents but is still very useful for humans 
It should take in an iface and you should supply what you want
It is not. Forking a repo just to change one file is not worth it IMHO. It could be just a .patch file
i forked a RSS kde widget today actually. i took the easy base and added new stuff to it to make it work better for my usecase
Except that this patch file will get an history over time
A history tied to the upstream repo
Well, the change is straightforward. Between stable crypto and noble. I already used noble but upstream uses stable

Unless it's really dumb and in that case PR to the upstream a change to use an interface and use that interface in your code
Then PR an interface because in that case they should be using one
It does not align with upstream
You shall inject your crypto library into it
that's far from the entire website though 
Then upstream is arguably bad and you shall fork it
And no, I will not maintain a code 
its about 70% that contains AI. there isn't a lot to the website all in all
I liked the idea but kinda went too hard on AI from what i heard
I will not be a good maintainer for the forked project
You should not, you should not hard fork, only soft-fork
Just maintain a downstream
just fork it. for forking it. i have a bunch of forks i never use 
Also actually i'm looking into other system than git
Patch based
Because patched where the true ways
DARCS or smth like that seemed interesting
In the theory
pijul 
i'm currently looking at gitbutler.. what makes it unique? compared to just github desktop or gitkraken?
Wait no it wasn't darcs
Parallel branches aka working on multiple branches merged into a virtual one
i barely even use branches...
I've heard that name
the way they handle moving changes around is also quite nice but i don't remember how the other tools do this ngl
That i read there
https://en.wikibooks.org/wiki/Understanding_Darcs
Darcs is a distributed version control system. Being a distributed system, the darcs model is rather different from that of more popular systems like Git or Mercurial and these differences take some time getting used to. One of our goals in writing this guide is to help you get a better grasp of the darcs model. We also hope to provide a usef...
Sinner
i just push to main
just let me drag and drop

Bro works alone

Y'all are too easy to fork some dependency 
i worked on like 10 man teams. we all just pushed to main 
Pretty sure you can cherry pick in github desktop with drag and drop
individual hunks too?
Tbh i don't get dependencies
Like i don't use
the only time i use branches is when i need to just test something out or i have an idea for a feature or a rework etc.
Forking it means vendoring it and keeping up with upstream. While maintaining patch on your own project makes it clear that you do not intend to vendor it 
i've grown to appreciate the way meson subprojects work
Me when i just see a message whne repo change version do quick git pull, rebase my custom branch on top of it and push
Done in one command
Then even more so on native as adding deps may make or break your apps. Especially if you compile it for specific version
we had per-ticket branches, a dev branch, an "old version main" and a main
And when it doesn't then i edit the patch, update the meta branch keeping track of the patch evolution and still in one command
i never pushed to main. mostly merged. sometimes force pushed 
older colleagues never getting good at git
You shall not use dependencies unless truly necessary is my mantra
Now imagine if we use submodule and their ergonomic is not shit. Everything is done in the same repo 
working with the code from several year old cuda projects does mean a lot of patching to make it work with newer versions tbf 
Works quite well
I mean the idea is good except that the way submodules exist just cannot work as it
It's just horrible dealing with them
Vendoring is unironically the better git submodule
Exactly why I complained I wish submodules are more ergonomic 
So we don't have to happily fork for every single small changes that we need to a deps 
They can't really be
one thing that bothers me with vendoring is that it's often very hard to figure out wtf you changed or if you changed anything at all
That's why you should not make changes to the vendored version
You should instead add a patch on top of it

And also keep track of what commit is that vendor
See? 
And also use git rerere
Oh no i don't mean a patch file
I mean a patch commit

And then when you just move that commit and rebase it after pulling in the new changes git rerere will fix shit for you
why would you make people search the history over just having the changes in a file
Because it's way more convenient
I bet it would have been easier if you just have patch directory on your project repo, and the submodules is stored on vendor directory 
all roads lead to meson subprojects 
But then you need to apply patch each time you build
Which is bad and useless
Git's whole point is to track changes
I'd argue forking is even worse 
so why keep a patch file instead of committing it
I swear i'm not crazy
After having to maintain soft-forks cause i disagreed with upstream i just started finding "simple" solutions
all roads diverge into many ways of doing things that cause chaos and bloat and suffering
i'd say if you need major changes then just vendoring makes sense
Correct
but for simple patches i'd personally rather just commit the patch file than the entire dependency
I have no qualms with forking and vendoring
patching is virtually free anyway
Anyway you should not have dependencies anyway
Just put that left-pad function in your own code
And use a language with a better std
unfortunately a lot of useful software is written in languages with no/shitty standard libraries

can't really do much about that
Because we're all xy'ing here, the initial problem has always been the language was missing something
c++29 is going to solve all problems trust 
That's why i like Java
It's arguably not the best, but it has dam good thing in the std
Once C29 drop i'm not using C++ anymore cause C will get defer so i don't have to use goto error anymore
is this where im supposed to sell you zig now
No i'm not a fan of zig
Sorry
Nor rust
They both have cool concepts but not a fan of the languages for different reasons
I really just like C no offense to others
The language is missing path planning std algorithm 
Or bloom filter 
Or argon2id 
Or TLS ciphersuite 
random little fun fact, the cargo lockfile can contain duplicate entries and cargo vendor understandably loses its mind over it
just remembered 
this used to cause a bunch of problems when building gitbutler with nix but seems like they fixed it upstream by now

iirc the specific issue was that they were using git dependencies for gitoxide but something also depended on it through crates.io (same version and everything) 
so it didn't know which one to pick
Do not pick go back to assembly or you can't do that
I miss danidev too much, i know he dropped banger game but i miss vids about karlson and milk
In other news, cuda now support writing the kernel using rust via cuda oxide 
Is the Nvidia A2 a good bang for the buck for LLMs?
Depends on how cheap you can get it obviously
It has very, VERY tiny compute though, it's apparently comparable to a GTX 780
oh i just found one for 495โฌ
Ain't no way it is worth that much
a friend is looking for a gpu to run 20-30 b llm models for max 600-700โฌ
Get 7900 XTX from SamVanTriangle for around 650
24GB VRAM will be plenty for a 20-30B range model
Apparently he can't put in a gpu that needs power from the power supply because the power supply has no more pcie connectors lol
In that case idk if that even makes any sense. Or he needs to get a new power supply first...
Well you're out of luck then, no LLMs for you
As far as I'm aware, there is no 24GB VRAM GPU that does not require auxiliary power connectors
I am apparently correct
oh well
Get a PSU with more PCIe or adapt a CPU power into PCIe power, then try again
they're missing a lot at the moment but hopefully that'll change over time so I can fully abandon c++
Picking it up on Thursday BTW
The owner made the nice gesture of installing fresh Windows 11 on it (which is obviously being wiped)
IG is good enough machine for having coding agent sit in it
Maybe SSH via phone to tell "ok go do that" sometimes
nvidia's L4 does but it's expensive af

24G of vram without a PCIe power connector 
also TIL there's an sff version of the rtx 4000 ada that falls under this too sort of but 20G of memory
I'm sure amd has some obscure gpus like the l4 too but I don't feel like looking into that 
not that it matters, they should just get a new psu

Do the m40 or p40 use external power? They're ass but they're 24gb of ass
M40 is a 980 ti with 9% more SMs
it's not going to run on slot power 
P40 is a 1080 ti with 7% more SMs
AI agents are advancing research-level math. ๐
๏ธ๏ธ
๏ธ๏ธIโm thrilled to share @GoogleDeepMindโs AlphaProof Nexus - an agentic framework for formal proof search powered by Gemini.
๏ธ๏ธ
๏ธ๏ธWhen applied to a set of open formal math problems, our agent autonomously solved:
๏ธ๏ธโ
9 open Erdลs problems (including two open for 56 years!)
๏ธ๏ธโ
44 Online Encyclopedia of Integer Sequences (OEIS) problems
๏ธ๏ธโ
A 15-year-old open problem in algebraic geometry โ
A 7-year-old open question in min-max optimization
๏ธ๏ธ
๏ธ๏ธWe are collaborating with mathematicians across disciplines - from combinatorics and graph theory to quantum optics. Ultimately, these results show the massive potential of even simple agentic loops powered by Gemini.
๏ธ๏ธ
๏ธ๏ธRead the paper here: arxiv.org/abs/2605.22763v1
HOLY
looks like the answer is zokor, funky name 
I took codex version of chatgpt 5.5 to spin and now I have this game project that is roguelite / roguelike card game lol https://puu.sh/KNBrN/e59b5f1a7a.png
Hello everone I am having a strange issue with my GPU, I have a NVIDIA 3050 with the most recent driver update installed, yet the game Forza Horizon 6 refuses to use the card it seems, the game will start at 90% usage and 75FPS but over about 15 minutes it will go down to less than 30% GPU usage and will get less than 30 FPS. I really dont understand whats happening and when I look in task manager nothing is hot infact, my GPU wont go above 50C, any ideas?
usual suspects would be... driver compatibility issue (they get new bugs and stuff) or card being faulty
if you can play really heavy to system game for hours without performance issues then it's probably not a card... if it utilizes same fancy features anyway
it could be bug with the game or driver
I can run beamNG at 60 FPS and that game is quite taxing so I'm hoping its just Forza's optimization feature killing my frames and a update will fix it soon but I'm not sure
I dont think its driver issue because I had the same issues with my old drivers before updating to the new ones
CLORD IS A PROGRAMMER????? 
Damn it really is expensive
I used to be really passionate about programming back in the day but it was like visual basic, qbasic, basic... I like to design games and my coding knowledge from those days is really outdated XD
anyway, the game is called "Spring Neko Crystal Heists"... I made a page for it over that itchio place... I gave the game soundtrack
I haven't noticed a problem like that, but FH6 does have some graphics API usage bugs that proton/DXVK need a workaround for.
what power supply
damn, gpt 5.5 codex made that look good for how basic the designs felt
check with gpu-z for also something like memory temp and hot spots
Would be nice if the devs fixed their game 
find it funny that agent loops only has 2 possible path. either spewing up complete existential crisis nonsense in the end or spew out some kind of these new solutions (well, formal mathematical definition/solution) to some arbitrary problems
apparently was solved, just formalized
chatgpt used to struggle code anything competently and stuff... codex chatgpt 5.5 barely makes mistakes... issues have mainly been pointing out some improvements so stuff like how much armor playable character has isn't mentioned in two places of the interface as it's redundant, etc...
yeah that's ai for you
this will be a fine addition to my collection
I need alternatives to
and
for Teams at work. So far ๐ซ and ๐ have been filling that role.
ok so update, it seems all I've had to do is turn off my pc's bluetooth and now the games runs at 70+ FPS constantly
no clue how bluetooth was the cause but yea thats what everyone was saying and sure enough 
well bluetooth is like wifi
but not as optimized because not everyone uses bluetooth
you think it would be optimized for the pc version when the game specifically tells you to get a controller for a better experience
bluetooth sucks because the max bandwidth is <2 mbps, it's the lowest priority 2.4 GHz traffic, and it needs to constantly frequency hop. No idea why it would cause performance issues in a game though.
yea this is the first game where I've had this kind of issue but many others have had a similar issue to me and for somereason bluetooth seems to be the reason why my game was slow
thank yall for the help!
Vram?
Oh
Bluetooth????
What the fuck
bluetooth is literally balkan witchcraft
what?
you will then one day run into an extremely horrible and unexplainable issue all at once
nobody is above the curse of the blue toothed wizard
I surely am 
my experiences with bluetooth have been "why is the connection dying every few days" to "how did that change i made fix it" to "why is my connection dying every few days" to "oh the connection died again because the bluetooth drivers are stupid, time to unplug and replug the bluetooth antenna to fix it" and has stayed there for some inexplicable reason
nothing i tried after that ever fixed it and it's currently Good Enoughโข so i just leave it be and it works mostly
still would absolutely avoid using it in the future because fuck bluetooth
the only issues ive had with bluetooth were just bad wifi board
just replaced it
literally took 20min of troubleshooting
my board has bluetooth but only for windows 11 apparently
@stray dragon surely if we sacrifice sam to the bluetooth monster all our issues will be gone because he seems to have no issues
and i do not feel like going through that whole process
true
works on his machine
maybe we can steal some of his powers
I think the luck in in the triangle necklace
no will give it to bluetooth monster for better favors like better ping or permanent cheap deals for computer parts
much better than dealing with whatever the hell is bluetooth
permanent??? you ask for too much
his luck is permanent too
law of equivalent exchange
even though both are permanent there is a great difference in the value of them
like how there is a difference between unlimited bacon but no more games, and games, unlimited games, but no more games
then how about converting his specific charm in no bluetooth failure to a general purpose with 80% success rate at luck with computers in general
surely thats not too much to ask for
i think this will work
works on my machine means everyone should just use my machine
i will go consult with the bluetooth monster
i'd rather deal with live human sacrifice and the aftermath associated with it than subscription fees
how dare you.

the only issues i have with bluetooth are just stale handshakes when i swap linux <-> windows cuz i cba to share the pairing keys between OS, so even tho they both are paired to the same headphone or something it'll not work with one since the headphone is expecting a different key from the same bluetooth device as previous
I don't really have any bluetooh issues
even with what i said above i dont have the issue most of the time since they haev so many pairing slots now that they just save 2 pairings
inclusing using my linux laptop as speakers for my android device which somehow is just baked in???
the AX200 that came with my mobo has a manufacterers defect that makes it fail with environmental power-saving setting
yeah my pc loves to show up as speakers for my phone
kinda annoying lmao
but working
t we have more sacrifices t bring out the room of the souls of the damned
i keep using the shortform of "environmental power-saving setting" by accident so i keep getting automodded
hi mods 
they dont want my soul, they kicked me out where you think icame from
finally, it's been 9 months since the last use of the room of the souls of the damned
surely if the devil dont want you the heavens will be pleased with your head
neither
wish we got more opportunities to use it
highly unlikely it would be that case
strange, I've already been rejected once
but you know how it is finding proper sacrifices in this damn economy
well this your more opportunities let's get going
on it
never underestimate my capacity to annoy everyone everywhere all at once
can we sacrifice the mods?
the alter will be dusted and polished until it shines and the smell of blood will be almost removed
you do realize the gods have a shorter temper and will be off with your neck the moment you annoy them too much
except the jester
yeah and thats me not you
who will kill you on sight due to being more funny

uhhh get the clorox and some cleaning acid too the stains dried too much
get some blacklight too
talking to person, mentions jester
looks at name
jester i've tried everything and the smell has never been fully removed
it's just one of the charms of the place
we're sacrificing YOU (not the jester) NOT ME (actually the jester)
fuck it man throw in some scented candles or some essential oils or something the summoned guy was complaining about the smell last time
said he'd tank our reviews or something
idiot
uhh sorry about that, I have been sacrificed so many times of my own volliation the gods are no longer allowing me to do so
bro that was the sacrifice not the devil we summoned
im talking about the guy who took the souls
you WHAT
no bad reviews
- it's a great trick to get the next guy immediately to owe you a huge favor
bro are you the fucking reason there are hood wars in hell now
it's hell who gives a shit
they said the buff to the the invidual I gave was so powerful that they had to nerf me from constantly just using it as a permant buffing alter
man fuck you mean who gives a shit we do our business there
yeah but this time its not of your own volition
complain any more and i'll teach them what weaponized baby shark is next
Idk, they just made it so that I'm immune to all holy weapons
we're summoning the devil to sacrifice some programming members
we're sacrificing triangle for better bluetooth
dumbass it goes in REVERSE
who said anything about holy weapons
you sacrifice them to summon the demon
Average low level programming discussion
YES YES I KNOW but the normies dont know that shut it
fuck you're right
So low level we're at the river styx
I finally gave Zero proper tool call/some self programming capabilities and the first thing she did was write a python script that launches a rick roll on youtube. She's truly mine.
yo bro where did you put the razer laptop i gotta code the summon script
alternative link i hope
i sold it for gas money
you know how it is these days
I gave her the link, she asked for it.
MAN THAT WAS THE ONLY LAPTOP FOR SUMMONING HELL
So she didnt surprise me, but the idea was hers.
buy a new one
fine ill use my hinge problem laptop for now
do you have any idea how expensive the program is to even get any copy of???
even the pirates sell it for a profit
I keep telling him he dont listen
Nani
all he sees is the money we make
fuck where'd i put my boombox and cassette full of wh40k tapes
I think somewhere in the summoning room maybe, as long as someone didn't misplace the key for the grand hall
i knew your ass would do this so i cast clone on them to keep another copy
this "clone", what else can you cast it on
no we lost that long ago we just took the door apart
as long as its not too big and not currency or valuable because that needs a bigger sacrifice to unlock the spell for that
400oz gold bar?
I don't think there are any rules about valuables in terms of infomation, those gods have still not patched that exploit
no like I said it cant be valuable it needs a sacrifice of a head of state
the heavenly version isn't patched but hades fixed it long ago
as for info
yeah nobody patched that yet
why is there a heavenly version of clone anyways
you could try cloning bitcoin
too many hashes, no point in trying, Can't randomize into a valid hash
different network architecture
even if it is cloned it is effectly entangled anyways
Ah, for when you need to cross world lines and need to clone entire cities and other templates
wish we had templates
instead we can only exactly clone an object as it is
mac for gcm calculated over ciphertext; i thought it is cryptographically insecure...

Hmmm thats why im now running proxmox with zfs on my main :D (yes im aware snapshots take storage
Rewrite it in luau
God damn message was old
My phone just scrolled for like a entire minute
Hi, im now in the now again, my phone decided that
how to compile lua to wasm
apparently this is enough to fix my type issues... actually now that I think about it why did I not think of that sooner
Idk I'm not a web dev
nah, everything should be rewritten in rust
rewrite it in c++
fuck no
rust is a cult
change my mind
nothing should be rewritten in c++
c++ is fine
safe rust is memory safe language
you still getting logic errors and blow up on any unsafe that violate contracts
but unlike c++ you have:
- better stdlib
- standard build system + crates ecosystem
- unsafe keyword that highlight contracts
- remind me what was fourth thing i was thinking of
i not remember actually
better stdlib and unsafe contracts > c++
only thing that c++ do better is generic coroutines
you just cant do non-standard coroutines in rust
yes
ignore this ppl who want you to rewrite everything, that already good enough and work, in rust
but any new code, or complete rewrite, should be done in rust
there literally no reasons to use other programming language, except for maintaining previous code
pyo3 for rust is actually quite nice
in 2026 you have everything you need
yeah surely native bindings exist
you not even have to do ffi
Y'all speaking about sacrifice meanwhile this is Eid al adh where we actually sacrifice lamb and cows 

Rewrite in ANSI C, the only true compiled language. @glass jetty
Agree, it doesn't even have a compiler spec sheet afaik
And, ive yet to see a actual rust project that doesn't use .unwrap or spams unsafe
(and FilC exists and is just a quick mockup)
claude sonnet 4.5 my goat helping me with some typescript typing stuff
bweeeeee
it is the part of language btw
claude and vedal ai being the only useful ai i need
one for code the other for fun
like neuro, low latency causing it to spit out stupid stuff
lmao
๐ญ

its definitely faster to convert it to binary and then do a right shift of 20 
I have no idea what study mode is but....
https://m.youtube.com/watch?v=nx_zDqrMG8c
that's some quality shitposting m8
or i'm happy for you
or i'm sorry that happened
yeah I know my friend gonna give us some meals hes supposed to cook the meals and gift em but hes being lazy this year
The alternative is giving the raw meat 
bro he wont even make it raw dude dont even wanna do the butcher work either
thing is going to be handed to us live
Hey, you got to learn how to do it yourself 
my uncle had the misfortune of doing that when they once shipped a live goat to him out in the forest instead of rations while they were on an op 
parachuted the damn thing into the field
A goat experienced skydiving before a turtle. What a world we live in 
what a GOAT ๐
me rn
getting headache trying to read tokio code while lsp cant provide support since he not see features being enabled at all
why i bother
why not just download tokio repository itself and enable them, surely it should make fucking lsp work
can't you enable features in a .cargo file or something so your lsp knows what features to look for?
Come on, do it. Fork the repo 
without enabling features you cant use them
lsp just saying fuck you and spawning another PROCESS in another root (tokio root)
i can spawn like 100 rust analyzer processes this way
for each crate i have as dependency
wtf
100 processes per dependency crate what?
id here is lsp server id, that nvim assign to lsp process
one lsp per one crate
ah ok sounds more reasonable
i just open my projects like normal and it then just works.
aaaaaaaa
why is spectacle not enough?
cuz it only do screenshots?
it redirect to obs or whatever
what
ok they implemented own backend i guess
i tried to use it long long time ago, and it offered me using several backend variants, one of which is obs
i wish it would use obs tbh
the mp4 backend they have is absolute garbage. i record it in vp9 and then convert it after saving automatically to mp4 to have decent quality
๐๏ธ $1500 OFF ANY Online Tech Bootcamps. See if you qualify for the JOB GUARANTEE! ๐ https://piratekingdom.com/SPRINGBOARD
โฌ๏ธ 100% FREE until you're hired! 1-on-1 career mentorship from top-notch professionals at FAANG ๐ https://piratekingdom.com/deals/pathrise
โ๏ธ Get promotions and pay raises faster! ๐ https://piratekingdom...
These benchmarks have been specifically approved by NVIDIA, as being workloads where Vera is competitive. It was forbidden to publish other benchmarks. There are plenty of other benchmarks, especially for scientific/technical computing, where Vera would not be competitive with x86 CPUs
how many times does nvidia have to be sus
also that means the nvidia vera cpu are shit at hpc
i guess they're meant to spend the rest of their lives feeding rubin's tensor
@glass flower
and they both handle file change/ file save events
i had up to 5 same msgs at some point
usually its 3/4
cuz i use nvim as editor with lsp support, not as full ide (i mean i usually not have 4 tabs with 6 split each)
i also use nvim

but also isn't this expected behaviour? they are different crates they can't share the same diagnostic info
when you open tokio from your crate
it looks like this
1 moment
rust shares all the instances... but ofcourse copilot starts a new one each time
the autocomplete and next edit suggestions
i used to use copilet before agents and chats were even a thing
i remember
Hi

i was in the closed beta 

old vibe coder
even worse

I just use Claude free model
wait i see a difference... why is your rust analyzer setting the root directory. instead of using workspaces?
vibecoding has not existed and wasn't even a term when i started using AI
i used same as from original docs/guide
https://charaverk.online/git/CharaVerKys/nvim/src/branch/master/lsp/rust_a.lua
why not use the nvim-lspconfig one? https://github.com/neovim/nvim-lspconfig/blob/master/lsp/rust_analyzer.lua
because it is plugin?
its a collection of settings
you can just copy the single file
also whats wrong with plugins 
being plugins?
What are u guys building
Why would someone use neo vim over other ide?
customizibility for me. it just flows nice
Iโm just used to vs code + GitHub copilot
At most Claude chat bot on the side
Like how u would use ChatGPT
tui + light + free + opensource + lua + big userbase
- possible c bindings
- i understand how it works, well, partially, at least
and its all assuming by ide you mean something that can have vim-like movement
another vibe coder.

What tua and lua

Whats vin like
you move your cursor using hjkl instead of a mouse
The other thing like nano?
Yep the text editor in terminal I see
w/na$0kjhlKJeEbW
correct vim/neovim is in the terminal 
The future of coding in Gen Alpha is in good hands imo
Neuro will be our next Claude
Wait and see
Gen Alpha is gonna be completely screwed 
Soon neuro will code better than my Claude
ts so over
we are cooked
programming is dead
but not because of ai being good
I wonder how will millennials Gen X and Gen Z be able to hack Gen Alpha system
Neuro isn't even years close to Claude opus in terms of intelligence what are you talking about ?
Joking
I donโt even use that
I use free model lol
correct. neuro is better 
On my own pc
Tui and Lua is what

The free tier sucks anyway 
Enough to build stuff to impress ppl
At least it can run the stuff I want
Wait isnโt tui just cli
terminal ui
Oh gui

U guys use gui on terminal? Called tui?
not.
no
Ok lua is just another language that neo uses
no
TUI = Terminal UI (Stuff like ncurses) != GUI
no
Lol I guess I don't need to bother
btw
i found
handle for tokio is thread local
it is not waker data
-# as it should not be rawwaker data
only took several jumps with lsp
you are not
you cant rewrite this in rust and make it work
it is absolutely ill designed and ill programmed
but it works 3 month from now without any errors
https://charaverk.online/git/CharaVerKys/vpverk
give your ai task to rewrite this, and try to run
and see if it even can do basic socket stuff
and its all is cli.
This is what you guys call keep errors in code that pops up months later so ppl can't fire u?
Who in their right mind uses Windows 11 
it's like an exam question of "how do you delete windows system files and convince windows to do it"
i currently rewrite it in rust https://charaverk.online/git/CharaVerKys/vpverst
cut i planned full rewrite anyway (using this wrapper, that, now, i probably will never use actually) https://charaverk.online/git/CharaVerKys/ats_socket
not use windows
best answer
What even is vpverk ?
why rewrite when you could just use somethinge else 
what are you even saying? i just said it work without ANY issue for 3 month of daily use 18/7
like what for example?
idk macbook 
isnt it in readme? or i not created it?
tailscale
what?
what?
bro confused by his own pattern 2 minutes later smh
isn't it a vpn tunnel?
There is no readme and no docs lmao
Ah description says
bare socket vpn via tun
ah right
cuz this is fucked up
i created code, made it work, and use it
no statistics, no passwords
wait rust rewrite

Not worth







