#programming
1 messages · Page 467 of 1
the peak of my raiding life getting cutting edge was the peak of my weakaura slop spam
so maybe that was the secret
Nice terminator
Guys, please pray that my car battery still works, its been cookin cuz it wouldn't start after sitting for uhhhhh months™ sooo..
boil it a few hours n'she'll come around
i showed you my terminator please respond

Accidentally boiled it for 18h, right until my charger was throwing its tantrum once they reach 14.8V.. sat for about 5h and stablized at 12.7V.. do have to add that it once dropped to 8V because sitting.. whoops
Its just a.. 40? Ah one
Smol 1.2L engine
But it just went click and all lights go off.. gonna try in a minute
If.. i can get the damn battery installed.
Fuel in it is likely also stale soo.. yay
(yes yes, very programming.. its a honda, so a laptop may be plugged in at some point)
technical talk channel not just programming
arguably car shit goes from kinda technical to very technical
correction: channel for all Toast matters
i cant remember what your exact gripes with wayland were but if i recall correctly nearly all are resolved in the last 12 months and it was mainly severe pain around remote video and audio piping + the whole headless rdp issue where you had to confirm connections on the target or something
If it can't Neurolings it's a failed standard
remote audio is almost there (tm) there's an open issue that hasnt quite been shoved through to allow OpenPipeWireRemote() method via org.freedesktop.portal.Audio d-bus
so that's the main holdout still
there is Shijima-Qt which should work fine for that
It not support all features everywhere
the wl_shimeji in question
lol there's this giant box around them kek
works tho
id imagine this is some weird transparency issue that is fixable but it is funny to look at like this
text is visible behind
pretty much made for Neurolings on Wayland 

kwin window support pog
Yo
Yo
Do you know anything about game dev?
what's a game
💀
me
It's edible
Ewww no transparency
So trash
there is transparency it's just set wrong currently
I have a question for you then
holy shit im haxxing
I am making a game engine which is a better thing to be already in the engine already made
repeat in english
incredible
wayland
electron
windows
kotlin
linux mint
neurosynth
I am making a game engine and what's like a thing you wish which was already pre made?

Yes
rendering
tht's it tbh
if I can't do something in unity and need a custom engine
then only rendering is a pain in the ass to implement
you'll regret having to handle input manually
input is chill
can confirm, works perfectly fine on wayland plasma for me
3 million edge cases
Its honestly not that bad
gotta love edging
Use a dictionary to do the mapping
mental illness
yeah
dictionary + library to handle it
and it is as simple as checking with enums
Dont need a library to handle it. Its built in to glfw or sdl
This is how the engine looks like rn it's 9 days old
those are in fact libraries
Its built into the window library, you dont need a library for input

my point is you use a library
whether it's the same one as the one you use for something else doesn't matter
my take on custom engines is
you either make it as a hobby project
or to handle something that generic engines can't do well
because from a game dev's perspective it's almost always just better to use an existing matured engine with a big community and a lot of resources
last I tried to make my own it was to build it around noita-style pixel simulation
but my patience died when I clashed with opengl lol
Mine is built to run on a esp32
Its not runing on esp32 yet like it will in few weeks
Different interpretations
we're both saying the same thing in a different way but you're disagreeing with me lol
I disagreed with needing one specifically for input
There is two toast
two 
3!!!
bro fighting a shadow
there's a few more than 3
hooray
There's like 10
Like i said, i misinterpereted what you said
Anyways, what was the original question even
works with windows
shimeji-wl declares blur on its entire surface which i feel is probably unintended
that was 90% of the setup
On what DEs tho?
Well of course
They specifically target KDE and then on other things it borks
they use the wl core
hyprland doesn't work as per the GitHub
neither does gamescope
it should theoretically work on other environments but what he said
the blur thing i think is also the issue i was having with shijima-qt
Additionally, wl_shimeji will never support compositors that do not implement the wlr-layer-shell protocol, as it is a core part of wl_shimeji's implementation, so no Mutter for now
looks like gnome is out too
It's funny to see toast toast and toast is typeing
how do you think i feel, this is so irritating 
Hi fellow hacker
yep, shijima-qt spawned this evil while wl_shimeji spawned the neuro
Cinnamon/WL?
Hyprland nearly works but that has that bug causing stretchy textures
vaxry seems to think it's a feature so it's not getting fixed
I am gana see is there another foox or any fox name
gamescope not working is kinda obvious, it's mainly an X11 compositor to begin with 
There is a lot of fox
who would call themselves toast
or fooxy
honestly
- doesn't really exist yet
- GNOME fork so nothing Wayland-related works anyway
Only Wayland I have
there's technically nothing stopping you from just installing kde
and then using it
live
reaction
reminds me i should delete kwin-x11
what are you developing that requires xorg except perhaps some sort of xorg based shrine to the x god
me
App that needs to be able to move around windows spawned from other processes
I should
fix my WoL

it randomly broke some time in the past few months
haven't investigated yet but immediately assigning blame to Realtek regardless

Xfce on top Xfce ❤️
Cinnamon
What is up with the heart emote?
Is the discord mobile app bugged again?
Would not be surprised
uh ?
it's a heart?
there's nothing wrong ?
All of my hardwork and it turns out a dude in china has done the same work as me 
so it relies on what wayland considers a security hole (and i dont really blame it there)
compositors can do this just fine on wayland tho
Wayland incompatible so me needs Xorg
it's just a standard unicode heart that discord display as a twemoji heart
like it looks normal to me
Doesnt show up in screenrecording so idk. Weird graphical glitch
very weird
Also me no deal with fragmented compositors
hyprctl, swaymsg can do it on tilers
for you in particular since it's related, gnome-shell can probably do it and more
Every compositor need different implemenation
actually, ext-zones just got added as well
which is agnostic to compositor
i am struggling to find the specific usecase that requires unconditional control of the position of all on-screen windows at any time without regard to the process that owns the window that isn't a massive security
just asking to explode in some way
Wayland weird
Meanwhile literally every other OS and windowing system
macos doesnt allow it except with explicit per-app permissions that you must grant ahead of time, android/ios dont allow that kind of control (obviously), windows allows it but only sometimes (non-elevated process cant do anything to an elevated process owned window as an example)

x11 gives you a ton of freedom yes but that's for better or worse, usually worse
android does give you control over this no? just have to explicitly allow and it feels like it keeps nagging you every day about it
wayland doesnt have to be as restrictive as it is, on the same thread of inquiry
mmmm not that i know of
not for a random app
WindowManagerService owns everything
you might be able to do some shenanigans with accessibility services
you can also draw windows over others i believe but you cant just move another application's window
that might be what i mixed it up with
if you're rooted things probably change if you want to open up the same security spookiness
Glitch
‼️ ‼️ ‼️ ‼️ ‼️

x11 though i mean you can do literally wahtever to whatever else especially if you can get a sudo goin
if you can xinput test on the id of your keyboard you can read every keystroke everywhere
for any window with no checks whatsoever as long as you wish

XGetImage lemme just get a nice screenshot of that banking page you have open or maybe your bitwarden vault that's open
XTest imma just fake any input anywhere
anything on clipboard is universally just chilling waiting to be yoinked by anything that can execute on the system
most of this is not allowed on windows via uipi (and a other things fyi, MIC, uac blah blah)
macos blocks it with transparency/consent/control
a bit annoying sometimes how the wl clipboard works in that it goes poof when app quits but thankfully there's a workaround 
kde must have the workaround because that doesnt happen to me
ootb
also has the same security oopsies im sure
the clipboard manager usually takes care of it
doesn't work when the app directly uses the clipboard though
like in the case of codesnap
they found my top bar

The WebTransport API provides a modern update to WebSockets, transmitting data between client and server using HTTP/3 Transport. WebTransport provides support for multiple streams, unidirectional streams, and out-of-order delivery. It enables reliable transport via streams and unreliable transport via UDP-like datagrams.
codesnap
It's baseline
the vscode extension?

a bit of a pain
#!/usr/bin/env bash
printf '%s' "$ZED_SELECTED_TEXT" | codesnap -c \
-o clipboard \
--file-path "$ZED_FILE" \
--has-line-number \
--start-line-number "$ZED_ROW"
this works though, just need to invoke as task 
ooh two more fun ones, any X client can just XKillClient to any other x client without permission
XWarpPointer moves your cursor wherever, also no permission
oh it's out 
man how often are you sharing screenshotted snippets of code that requires them all to be prettified and is worth a whole macro
often enough
fair enough
i used codesnap like one(1) time and was like but why do i need that tho and that was the end
you should try making macro's. i have like 20 now. they are so easy to make that i just make them everytime something takes like 2 steps to do
Yep, since march apparently
it's nice for mobile
i have various alias and random macros for some things but idk if codesnap would rise to that level for me
if i would actually use it i would 100% do it 
the zed version is kind of unreliable, idk why i don't feel like debugging that mess any more

every time i use zed i end up in this weird no-addon zen state
i like zed
beyond like, a theme
is comfy
everytime i try to use zed i just am like "
i don't want to port all my vim keybinds"
god im so glad i am the opposite
i just do <space>q to close vim
well... that i do it is a lie. its because astronvim has it like that 
although it would be fun to be an 31337 vim haxxer for a day and just fly around some weird shit like helix or something for ffunzies

i like lunarvim in a visual sense, with all the stupid effects maxxed out
i just like nvim. it does everything i want and its really easy to customize
Helix mentioned 
i tried to use helix i really did (5 seconds)
im just not built for modal text editing
you just never tried it properly
ive done the tutorial
the first 2-4 weeks are a pain to do anything. then it gets a bit faster
and after a few months you just get used to it
i haven't done the vim tutorial yet at all
and i've been using it for like 8 months now
i feel like the best way to learn something is to just throw yourself into it and use it. and then when you need a specific action like find a specific string or replace something across the whole file. you just look it up. That way you aren't just remembering stuff, you actually need it and its easier for your brain to remember the need than the action
but what do i accomplish with it that i cant already do with ibm style (later windows style) control+arrows etc etc
its just faster
Ah filtered I'm late for karaoke
vim fans need to be gifted mice for bday
they just keep insisting on going keyboard-only
so uncivilised
how so, ctrl+H = find+ replace
idk how that's any slower than the vim style
not that the vim style is not fast
Didcord scammer asking for my country and age 
What about emacs fans ?
it just seems similar speed, better if you're more used to one or the other
well find and replace was just a example.
its mostly just about the concept that everything is at your finger tips
there's only a few that i can really see a replacement not existing for at all, like change inside quotes (or (), or inside a tag)
umm maybe %
what about jumping to matching things? like matching {[( etc
(i need to learn this more
)
i personally also just like the multiple modes concept.
yeah i suppose that kinda doesn't exist but also just search the thing and go to it
i suppose if you're trying to mash these all together into a single command that you compose
but i am not that galaxy brain that i ever even want to do it
i mean.. vim motions is just a way to move about and most editors support some version of it.
and modal just confuses and infuriates me because the same combination of keys suddenly do different things
due to my pea brain i guess
that is exactly why i like it. i can have the same keybind do different things in different modes
it really isn't that difficult. you just get used to it
besides 99% of the time its just between 2 modes. visual and normal. you don't really do insert mode keybinds
but then it just comes back to for what purpose do i want to rewire my entire input paradigm (and break muscle memory with non-vim keybinding using things, of which there are many, especially if dualbooting windows), suffer pain doing so for likely at least a few weeks, and then end up in basically the same place
using a cool editor is a valid reason
i feel faster writing stuff. and moving about. instead of having to use the mouse.
i literally don't touch my mouse at all during coding anymore
also its more comfortable
i havent found a reason to touch a mouse in most any editor unless there's some addon i need to interact with either
i used to before i got more used to the ibmland version of the same bindings
i feel like the time to get used to it would've been then but the thing is that shares a lot of commonality with just basic UI and terminal usage in that environment which is where i learned to do anything related
i mean its mostly just about learning your editor. if you learned the ins and outs of your editor it doesn't really matter if its vim or vscode or whatever. its just about trying to become better accustomed to an editor
and nvim allows me to get good at something and also change things i don't like in ways vscode doesn't allow me
agree on that definitely
who needs an editor when you can voice command an llm 
for example i tried for a few weeks to get rid of a file-explorer and switched to yazi as my primary explorer. but then i switched to snacks's file-explorer. and now i'm back with neo-tree.
in vscode i would have never had the option to experiment even
all of my source files end in dictated but not read, afunyun
well that's fairly unrelated to the keybindings thing and more of an editor issue
well but keybinds are part of it
as its the way you control the file manipulation
and that is part of the editor 
its all just one thing in the end. optimizing your workflow
https://github.com/NVIDIA/cuCascade this seems useful oooh 
GPU Memory Reservation Library. Contribute to NVIDIA/cuCascade development by creating an account on GitHub.
time to really dig in to emacs

technically i think emacs is ultimately more customizable than even neovim
What is this thing for?
if you want perfect customizibility.... just write your own text editor 
automatic tiered memory management it seems
looks like a memory budget allocation thing
idk if the ssd tier goes through gpudirect storage, have to look 
big if true
ooooh
they do have a gds backend
can cudf
i need something to ingest gigantic dna sequences so could be useful 
idk never used cudf
oh it can via Dask
but also is limited
so that's i guess what it's cuCascade is literally for
neat
https://github.com/llnl/Umpire exists for non cuda
i see nothing wrong with that, look at pkgbuild it's just cloning and building a repo end of the day if it's not already a bin
OR
if unlucky
you could be a poor (dumb) sap downloading the wrong google-chrome-legit-cracked-exe from the aur and get pwned
4 (1.2)gb with my vpn..
Total Download Size: 1206.61 MiB
Total Installed Size: 4381.68 MiB
Net Upgrade Size: 131.99 MiB
that's just one cuda sdk update
glad i not have any cuda installed on my laptop
you are a wizard harry
oh is it actually 
me when i get to download the sdk on my ultrabook because the only thing i care about is headers
just code withut lsp 
Just code with 
classic man
damn if only there was a way someone else would already be able to do this for you, like a large corporation

it's not even related to the rocm build thing 
thankfully nvidia's download servers are really good at least
hipify-clang is required to run most ML anything because they write their kernels in cuda and then go "we support rocm :^)" by slapping a hipify on the build process
And that
requires full cuda
Lol
i definitely have never done that

since when does hipify depend on cuda though
it's just slightly more advanced search and replace
and even stuff that doesnt actually use hipify (it's native HIP) like ggml-cuda which has a valid hip source tree maintained on its own, you pull in any dependencies which obviously happens pretty much always and those themselves will pull in cuda as part of their build path, even if they dont need it half the time
hipify-cuda runs through clang's cuda frontend
which needs real cuda headers

otherwise it cant resolve cuda-runtime.h, cuBLAS, cuDNN, whatever cuPoop you need to build
there's a hipify-perl which i have no fucking idea waht that does
i'm confused, you shouldn't have to build any cuda code
hipify-clang just spits out a hip file that you would then compile using hip libraries
that part needs cuda to exist
hipify-clang sends ur cuda oriented source through libtooling and builds a cuda source tree and then translates all cuda api calls/types/whatever into hip
from amd:
hipify-clang is a Clang-based tool for translating NVIDIA CUDA sources into HIP sources.
It translates CUDA source into an Abstract Syntax Tree (AST), which is traversed by transformation matchers. After applying all the matchers, the output HIP source is produced.
Advantages:
hipify-clang is a translator. It parses complex constructs successfully or reports an error.
It supports Clang options such as -I, -D, and –cuda-path.
The support for new CUDA versions is seamless, as the Clang front-end is statically linked into hipify-clang and does all the syntactical parsing of a CUDA source to HIPIFY.
It is very well supported as a compiler extension.
Disadvantages:
You must ensure that the input CUDA code is correct as incorrect code can’t be translated to HIP.
You must install CUDA, and in case of multiple installations specify the needed version using --cuda-path option.
You must provide all the includes and defines to successfully translate the code.

so hipify-clang therefore requires CUDA in the version you need, and then the appropriate llvm+clang to build it

seems like vpn i wrote broke for the first time xD
i will rewrite this shit one day
idk what will happen first - me leaving this country or me rewriting vpn to proper api and structure
the 2nd one seems much harder
anyway, try pacman -Syu second time
speakatron lmao
i googled that and found a guy on twitter talking about metal
glad pacman does not discard what it already downloaded 
yeah, me too
oh
tbf 60 gigs of this is huggingface and its accursed desire to hold on to everything it has ever touched
uv and paru together are another 60 kekw
Nix does that as well
does any package manager not
anime protagonist
Idk if that is a German thing or not lmao
i write ope all the time
which is a northern michigan//maybe more broad than that but definitely michigan expression
kinda like a combo of "oh" and "oops"
Lol never heard that before
it is apparently related too all the norwegian immigration to the area but idk how
unless there's a similar phrase in norwegian
in speech it probably gets heard as "oh"
to most
if you try to go thru the same doorway or something as a michigan person you'll hear an ope immediately
if you ever hear it irl, freak them out by pointing at your hand and asking where in michigan they're from
bonus challenge: if they sound canadian, correctly call them out as a yooper
whats going on here
normal day at anthropic
Clawd ai
same incident that lead to me having to run lspci manually over two whole times earlier
like a caveman
the horror
i remember it like it was yesterday
it was today, but i have brain damage
not even rich people mode saved me
183416661174888888333333333333333333333333333333333333333333333333333333333333333333333333744085277777777777777777777777777777777777
well the enter key is pretty close to the backspace
was vacuuming out my numberpad because i was holding a monster and for some reason decided that physically seeing the liquid would tell me if it was in fact a recently opened one or if i had left it on the desk for 12 hours and needed to get rid of it
and that action poured 2 drops of it onto my num pad
and instead of cleaning that normally, i decided to vacuum it
i got a question. does anyone here have a github copilot subscription? i was wondering if its better than using a local model. since 10$ a month is fairly cheap if its really good.
i only use autocomplete generally but i felt like sometimes my llm is more than just stupid 
I accidently ran a vram heavy thing while another vram heavy thing was running 
I also clicked stop right away and I think I crashed the driver
it's maybe worth it if you want nonspecific access to most of the frontier models in a way that doesnt bill per api call, i havent used it in forever but it also used to allow calling models through a specific github copilot model zoo type endpoint that didnt have additional billing attached to it, so for example you could route something like kilo code through it
what you're seeing here is the result of vibe coding
You german?
you get 300 codex prompts a month, goes a pretty long way all things considered (just don't use opus)
copilot models tend to have a lower context window tho
Cuz like, i am, i never do "Mh"- only do "Hmm" or "Mhm" (although latter is more of a acknowledgement)
i have the student plan and it's good
tho i don't vibecode
i barely use it either as smarter autocompletion
or as search in codebases i don't know
i basically just use it to learn more
i find Gemini 3.1 Pro to be the best model i can access
and with the student plan there's honestly no limit if you're not doing the thing where you spin thousand agents in the background
Ja
My condolences
like we're at the mid point of the month and i have used half of the premium stuff which only come from using that
also the context window of Gemini 3.1 Pro is 128k and it's honestly massive even when i force it to ingest the whole codebase i'm working on it's like 35% full so it's fine
Could be worse.
At least I'm not fr*nch
fr*nch 
well i'm only considering it because i want autocomplete nothing more, nothing less. i don't use cli tools at all.
the most i do is like select a region of code and either ask something or let it change that.
i used to use supermaven a long long time ago before they got bought out, but idk if that still exists or if github copilot caught up, haven't used it since they introduced NES
currently using minuet-ai.nvim for autocomplete and llama.nvim for instruction based editing. and VectorCode to vectorise for more context.
but if i used copilot i could use sidekick.nvim to combine all these into a single copilot lsp basically
copilot was really ahead when i was in the beta for it. but after a few months when it was paid i switched to supermaven. then that got bought and i just stopped using paid llm's since then. after that i only used local AI's. currently using qwen 3.5-9b for everything
For smart autocomplete I use whatever Zed/VS2026 have for free
zed has zeta. and vs2026 is copilot
(i could even use zeta technically since its open weight. but idk)
Ah cool- gibt also paar von hier.. nicht mehr allein loo
Gemma 4 26BA4B is so good for local use its unreal chat
Its just smart enough for all the tasks you'd want from an on-device model, just fast enough to feel more responsive than cloud services, just small enough to fit on most PCs
I've been using it for OCR, web search and translation and its an absolute beast
does it fit in 8gb of vram?
cause if so I may put it on my server that has a spare gpu in it
its a moe so technically yes
Its big trick is that its a MoE model so you can fit it on anything with enough sysram at decent speeds
Just use llama.cpp and tune --n-cpu-moe incrementing by one till you no longer see the --fit message saying it needs to free up memory
ah, that would be a problem cause it's on a chipset pcie3x4 slot
In that case try Gemma 4 E4B
You might be pleasantly surprised
Not a bright model but its pretty damn good for multimodal understanding, has native audio input depending on your inference engine and does web search pretty nicely
Also has very low hallucination rates on questions it gets wrong
You should be able to fit it at Q6_K with like 8k context
If that's too big try Qwen 3.5 4B
Its less stable (gets into loops sometimes) but arguably as smart if not smarter and smaller
why is it redownloading gemma4 for me... so weird. has there been an update in the past week?
im testing gemma4 31b_q6 rn vs gemini 3.1 (for the lulz) on a small python task (refactor)
probably a very easy test to not totally fail at for any decent model but mainly trying to get a feel for how it'd do on something pretty trivial/day to day, myself and also claude as a judge, ill have results at some point but i mainly mention that for context on this making me laugh w/ claude's last review of the changes done by gemini
different quant? mine didnt rebuild and i hadnt touched it for a week or 2
rebuild
redownload
lol the diff is the entire file except one line ok gemma
New llama.cpp fixes merged in and quant updates yes
Rebuild llama.cpp or download the latest binaries if you haven't

now i wonder if it supports FIM out of the box or if i need to do stuff to make it work..
<|fim_suffix|>
building llama currently. i did just test it a bit and it seems to work with the qwen style fim things
but the plugin im using is not actually hitting the /infill endpoint
nah, i need yo build my own gpu obviously
invent your own boolean algebra
"convert to rocm, make no mistake"
Now the next logical step is to optimize it
honestly i have a claude code skill specifically dedicated to fixing venvs etc w/ rocm when it explodes
The fact that even with venv it can still explode is a testament to python package management quality
it's always the same shit and the same-ish fix it's just "find whichever dependency updated and sniped this environment with suddenly a full cuda runtime"
you wouldn't believe the amount of ways cuda can somehow appear in a venv
At some point we'll see docker-esque env for python
python with cuda is a fun one
ppython with cuda but you dont want cuda at all
except in the small amount you need to hipify
"Duck this package management thing, we'll ship the machine"
so a little sprinkle of cuda BUT ONLY ON OCCASION
thats the fun one
sometimes i just say fuck it , its slow as shit but its better to use CPU then my enviornment be blown up tbf
if you run the main .py by accident with the cuda there the entire thing detonates into a supernova
But some things do require that lil sprinkle
I feel this
because as soon as any of these things sniff cuda theyre like "OH YEAH NVIDIA?? HEY THIS ENVIRONMENT LOOKS MISCONFIGURED TO USE THIS NVIDIA GPU IVE NOT CHECKED TO SEE EXISTS, TIME TO FIX IT"
gg
tbf I have a nvidia gpu with cuda drivers so it only half nukes, the only issue is cas I have a 50 series card I have to use a special compiled version or it nukes
so
I have a similar issues

i cant even use prebuilt rocm wheels for anything because either amd REALLY sucks about their build target with rdna4 nightlies, or (possibly and) something about my machine is just weird & doesnt like them
ye
fuckin python
This makes me wanna just go write a docker -like thing in go or something for python
idk how hard it would be but surely not too bad
ehh fuck it may aswell ,
Would be handy to have either way
lol
the only time docker has actually been actually blessed in my eyes was when docker pull vllm/vllm-openai-rocm:latest just worked
ye
I was more thinking something to manage the enviornment and anti nuke it to some degree
first time a docker image has ever worked without problems for me i think ever
similar to nodes package.json or smthin
tbf I think all it would be would be using exsisting tooling and having a simple thing to manage it
EHh I will have a lil look if its doable ill write it and throw it on codeberg
or mshtin
or github probs both
tbf

what docker things do you run that don't work ootb
that's kinda the whole point of docker
to be fair
I don't think that's a docker issue
I think that's an "AI tooling is a clusterfuck" issue
beyond that i just dont wanna run your lil 200mb binary in a whole ass docker container that pulls and hosts a full fat ubuntu with gui
fe
I have decided to actually write that thing if it works ig that will be useful to you
lol
let me know on results 
the closest ive come to that is basically pinning a ton of shit in the uv.lock
tbf the main thing is ensuring dependencys dont fuck themselves
and ensuring it runs properly
best thing to do is entirely avoid any upgrades to packages unless absolutely sure
What ima do is write this in Go since it should be a half decent language for this, just basically make npm for python but with a spice of docker
cas
U can just preety much define
fully
what you want
and just be able to run it
and relying on nothing being changed in that config file, your enviornment will be managed and not touched by anything unless u tell it to
I cant type but I think u get what I mean
this reads like a pitch for venvs ngl and specifically uv in general hehe
Tbf I am just writing this cas it will end up benefiting me
lol
aswell
and also is useful
What makes go useful
if it works better than the venv that kinda sorta works if you dont poke it with a stick then what can i say except godspeed
Real question
go fast
ye
in this context idk
Is it low level?
Is it JIT
it's compiled
So it compiles to C++ like IL2CPP?
It can just compile itself if u pass flags
Interesting
but I beleive default behavior is to C++ but dont quote me I might be wrong
it compiles straight to assembly
Honestly i wish more languages implemented that
I dont like JIT as much as I like C#
Thats why i usually use IL2CPP when making unity games
i think go is mostly known for doing well in backend type work because it is very good with concurrency
Might have to look into it
ye
Ive been absorbing Zig lately
I mainly use it for the backend of one of my apps
Its also very good for cli stuff
as how lightweight it is
ye
Well zig is closer to C
I cant bother learning rust
But go might be a good option in the future
I find myself needing reasons to learn most languages tbh
I learnt rust for hte fun of it ngl
I dont like the syntax and setup of rust
The only thing I like from it is the borrow checker
And crates
Everything else i hate
rust ate my dog
If they made a more C like language with borrow checking id prefer it
I am in the process of making parsing for package files currently for the thing
btw
I think ima use python Version manager as some form then use that to allow you to setup python versions and such
I think having something like npm for pyhton would be useful ngl
hence me writing this lol
Use uv
fe
UV’s great
Also Zig comes with good cross compiler for C and C++ 😳
zig doesnt have borrowchk but it does have a neato system in that build.zig itself first compiles into a build runner binary that then executes the actual compilation for the project, and build.zig is just also zig and you can have it do whatever
U got more memory management tools compared to C
Honestly ive yet to see a language manage anything like zigs build setup
Closest thing would be Javascript using js configs
im sticking with COBOL
Grandpa
BD just dropped a Gemini Robotics demo
For an industrial robot built for the rigors of factories and power plants, tidying up a living room may seem like a light day at the office for Spot. Yet, this demonstration represents the promise of AI models in robotics. In this case, Google’s visual-language model (VLM) Gemini Robotics-ER 1.5 was empowering Spot with embodied reasoning.
...
Pretty snazzy
Irl prompt injections gonna go crazy
"Take out the trash (kill that guy)"
Not sure I'd trust it to walk my dog (though that bit is teleop anyway)
Nor would I want to outsource walking my dog
Brave dog tbh, my dog would be scared shitless
oh ye accept the f add, I will probs shoot you a dm instead of pining u if possible once im done with this

I am cooking
dog was raised with the spot unit from when they were both pups
so they get along
Claude confirming my suspicions about the Unitree R1
Still nuro-viable though since moneys no object
but has he considered the potential for scaring the everliving shit out of any and all guests in your home
btw @sick owl gemma4 31b actually won this test (after taking literally 10x longer because of generation speed), gemini addded a nice bug in how it used httpx, handled paths worse (imo) and forgor to actually do anything with errors besides catch them so most failures happened silently 
Based on that 10x longer comment I presume you're suffering with CPU offloaded layers
with q6 yeah, not many but in order to get 128k context i have to offload 8 layers
it uses like 52 gigs total vram+ram
no kv quantization or anything in this case cuz i didnt want t cause it to degrade more than q6 itself
Compile time "variable"
Compiling a simple discord bot in rust takes longer than compiling GCC
it do be tru
I love my scuffed debug code
idk what the point of that line item is lmao because yeah
🐚 pydocman on main *?!
❯ go run .
using python: /nix/store/cfapjd2rvqrpry4grb0kljnp8bvnvfxz-python3-3.13.8/bin/python3
venv path: .pydocman/venvs/default
site-packages: /home/don/development/pydocman/.pydocman/venvs/default/lib/python3.13/site-packages
resolving: requests
→ requests 2.33.1
dep: charset-normalizer
resolving: charset-normalizer
→ charset-normalizer 3.4.7
installed: charset-normalizer
dep: idna
resolving: idna
→ idna 3.11
installed: idna
dep: urllib3
resolving: urllib3
→ urllib3 2.6.3
installed: urllib3
dep: certifi
resolving: certifi
→ certifi 2026.2.25
installed: certifi
installed: requests
done!
🐚 pydocman on main *?!
❯ .pydocman/venvs/default/bin/python -c "import requests; print(requests.version)"
2.33.1
🐚 pydocman on main *?!
❯
well somethign is workign
i can fit in vram if i only give it 32k context & set kv to q8 looks like
I really wouldn't set kv to q8
i like the actual emoji from the terminal still being present
i was kinda curious to see how it affected it
Hurts reasoning models more than benchmarks properly convey
I just copyed my terminal output lol
I immediately saw formerly infrequent errors cropping up in answers when doing it
this test was 0 thinking budget forgot to mention
tbf Iv just got to write the acutal intergration to the parsing of the file and that hsit, also make it so you can run scripts and such using hte files
I cant type gg
Personally 16k context is enough for me
Most of my local tasks are only a couple turns at most anyway
I will either fucck my sleep schedule and ifnish this tongiht or tmrw when I am alive again
most peak dev enviornemnt
this round looks to have been 13k tokens, not bad honestly (no thinking budget reducing that quite a bit probably)

i should do a test with grug at the helm
tbf if this whole thing im working on works , I will try get it actually packaged properly
cas go is preety nice to package
idk what in high hell perswayded me to jus make my own thing to fetch pacakges
but its fast asf
so I will take it
the fabulous button
i just know it works with lmstudio ootb
it does
I just currently dont have any llm stuff setup
as I am on a nix install
and i Couldn't be assed to config when installing this
My sheer laziness showing up once again
ill dm u the codeberg repo for rn as I am going to sleep and I commited the unfinished code
I will finish tmrw when im avail again
as its 1 am lol
there is no drive more powerful than being productive on a random task someone else mentions would be neat, if one has something they really should be working on
tbf I have been overworking myself on the other thing
so an ecuse for a break is heavil appreciated lol
Also this might actually help people and gain some traction lol
Ill probs jus grab a domain or smthin for it
and just give it a simple install script and shit
anyways gn
goin a little faster this time
Are you on a 5090?
It exceeds my VRAM even at Q6
that was the plan to avoid without spending the $4k that 5090s were when i bought it cuz it was basically this or like a used 3090 if wanting to stay under 2k which is annoying
3k or rocm pain
i wonder how the intel version of this thing is doing
arc pro b70
remarkably similar in shape
oof the vulkan driver for arc is not doing fantastic
I heard the b70 described as fast but a complete mess
that would make sense given arc has been around even less time in a form that made sense to seriously optimize anything involving inference targeting it
all the benchmarks im finding either use vulkan or if competing against rocm/cuda, use SYCL idk if that's what it's supposed to be using for this per intel or some sort of bandaid lmao
So fucking real
Go 3090
Don't bother with the newer slop
This stream ends at the heat death of the universe lol
I wonder if the random number will even hit the correct value or if it will loop before then

Quoting BLCNYY (@BLCNYY)
︀
@flowersslop A Microsoft event where Satya Nadella points at a slideshow that indicates Microsoft Edge is the number one browser to download Google Chrome
I thought this was a real image
Cool code
he's thrilled with this result
very thrilled

reminds me of the automaton son that they dress like an idiot
I thought you had made some code {Sob Emoji}
you know the sob emoji takes less to type than [Sob Emoji] right
doesnt have the same oomph

My gaming laptop's net name is MyNameIsSkillIssue because of how much suffering it has caused me
Also good to know mobile now has the new emoji support so my name isn't a box
what happened to this guy
kicked?

Still here
oh wait nvm
lol it wasnt loading mutual server
the day vedal makes neurocord a thing is the day we will all rejoice
I was just in a different discord server {smile emoji}
Wouldn't be hard tbh
Beyond just hosting
this is neurocord 
NN gets offloaded to a single system and the rest of the channels share the other
oh I got it
ohhhhhhh soooooooouuuuuuuuulllllll
neurovirus
Viruses are not allowed
oops
I liked the mv for that
hello m—
Music : https://youtu.be/xUXuY75fFbo?si=7dm-i399URF7h9Ma
neuro channel at twitch
https://www.twitch.tv/vedal987
Here is my new animation!
Ok I made this way faster than last time, not only because I learn a lot from last experience, and also anxiety, I just like I am some kind of parasite which is gaining views from Vedal's creation, I mean, ...
it’s been a few years and I still cant get myself to NOT PUT THE DAMN SEMICOLON AT THE END OF PYTHON LINES WHYYYYYY
just don't hit the \n and you'll be fine 
Pov c++ dev

[0] 2026-04-15 22:54:45,277 - Main - INFO - Heard User: As Yuri has been flooring, I always felt very beginner when I saw this. And I asked her to Please respond as I was there. Or if I could reply to that, I would never say to her question. And she would love to hear anything about her parents and kids that do not happen.

chat does this sound like me at all what the heck

who is yuri and is her family okay


@amber fractal how do u add custom emojis in ur name lol
Define custom
Because this is not custom
Some madlad actually got it into spec under inflatable face
Ooh it's not in discord
That's why
I cant see it on my PC cause I'm still running Windows from 2022
lmao
No one will make me update to Ai Slop OS
yea
Still can't believed someone managed to add flooshed into the offical spec
even if it is missing the blush part
"trust me bro" still works
Trust

finally done
the to maintain the code inside Unity a robot-in a MAINTACT for your actual, and synthetic LIDVT 3D. You of your a MINE an AI VT a MICK you is a MCV Wooden VTIV, the M V your MTR your SPA, an AI VT A MINE a self-aware, a MA VTR inside the B is a PA your (### your M a hardware-on a Actual for the actual UVVT an ABSY your the E you are a viral you each other VT a semi-virtual VT a MIDI in an IV, influential VT a digital, both an actual VT a virtual Andrew (your a key V an explicit M the the an traditional VT all individual president-in a robotic VT all ":"the A, the conducted. accused AI VT a Wooden VT a Wood VT a your A M an:// Press VT aVT a.SMTVT avt, Arabia you in theCYou in:// said ( the; your,.swing and. Angeles, a self-://.awt, your exact VT all-in a influential your human VT a virtual VT (your, the acclaimed,.getElementById VT the published inthe A the.S an interactive VT all the:// \u0437\u0440\u0435\u043d\u0438\u044f in a virtual all-in exactly to a youpt an actual M, his internal VT all theCC a 'actual VT a 'a real your `real a VT a each and{FT a literal VT,his specific, You N V your exact VT a VT..ComponentPlacement (A A N your VT,VT (your VT you in-cube you in a you you you in a you you in you-in a
oh my fricking hell...
what happened to this poor chatbot
ah found it haha
asahi linux for m3pro when
they're getting closer
actually it's about time that they release another progress report
those are fun to read
i mean
it's not syntaxically incorrect
Can we get Neuro emojis into the official Unicode table? 
Good luck convinving the boomers running the board
ignoring half of the actual usage of this emoji is how it managed to exist at all
just not mentioning the actual usage and making up some BS to include it
Make a pull request and hide it under the name of some other emojis 
what is the box in ur name supposed to be
Curse the pdf not giving me an image
which is very much one of the emojis of all time
flooshed?
Ye flooshed
i have this
hello brodie
ignore tabs plz
Offically released September 9 2025 in unicode 17.0
Konii is going to kill me
How do i find it with text input
I haven't seen it yet
The emoji
only just added it's existance on mobile like today
so still is going to take a million years
Oh so its not added as text yet
Yep
not there on tahoe though

erm
erm
I sure wonder
don't ask why i can't download it
i have only like 18gb of free storage on my school laptop
Lol my Claude never denies anything related to pentesting or Ctf 
it was yelling at me earlier for another thing I was testing
Haha
"I'm gonna stop you right there"
seahorse emoji when
Wait is that it? 🐴
No wait here it is 🏇
No wrong again.
This is the seahorse emoji: 🎠
"i need to impress my ta bro please" "aight"
Internet Protocol Version 8 (IPv8) is a managed network protocol
suite that transforms how networks of every scale -- from home
networks to the global internet -- are operated, secured, and
monitored. Every manageable element in an IPv8 network is
authorised via OAuth2 JWT tokens served from a local cache. Every
service a device requires...
bro no one is even adopting ipv6 who's going to adopt ipv8 atp
i thought this was an april fools joke 
I didn't look at it but someone in another server said it gives ISPs more control and is just an extension of v4
it basically is 
bit late for an april fools joke
it's a draft submission by some random guy associated with a newspaper in Bermuda
I wouldn't take this too seriously 
ive read that, the point is that it's reverse compatible with ipv4
Fr
I don't even know how to connect to an ipv6 ip address
North-south security -- traffic from internal devices to the internet -- is enforced at the Zone Server egress by two mandatory validation steps. First, every outbound connection must have a corresponding DNS8 lookup -- no DNS lookup means no XLATE8 state table entry means the connection is blocked. Second, the destination ASN is validated against the WHOIS8 registry -- if the destination prefix is not registered as an active route by a legitimately registered ASN holder the packet is dropped. These two steps together eliminate the primary malware command-and-control channel: connection to hardcoded IP addresses without DNS resolution.
eww,that
yep, more control...
Someone's trying to kill DoH I guess
Someone is trying to kill non datacenter hosting mainly
Meh, I have https sites running on my home network. The only really annoying part is the reverse proxy setup.
You become your own ca or let's encrypt the dns? Because damn, I am too lazy for that
Goodmorning 

Toast! 
Toast
buy domain from cloudflare
setup dynamic DNS
setup ACME to renew *.<domain> certificate
setup haproxy to forward the http domain from the router to the correct port on the NAS
Lots of trial and error
I WAS TALKIING TO GEMEINI ABOUT LAVALAMP API AND IT BRANG UP MY PROJECT LULEELELELEL
egosurfing
Goodmorning konii 
I haven’t touched normsync in 1 year
Apparently the api is at swarm.vedal.ai, I don’t believe it
a quick dns lookup confirms this
Yeah it confirmed it themselves lol
We NEED to find this api
The tuya dp
Wanna help me research and create a api for it
Cause vedal ain’t giving us one
Nah
ok
wait its you
Does it just cut data cables 
Oh wait but it needs them for some new smart charging or smth like that

no
Why so many non toast 
No toasts besides the DLL file
true
neuro, kill them all
Why not?
Is windows defender exploit real? 
https://cybernews.com/security/second-public-windows-defender-exploit-released/
What did they do to piss off the security researcher

fun zero-day
Where can I get neuro sama voice?
I wanna try testing
(ig let try make one myself, cus I wanna farming github repo)
I just use Caddy. It's super easy to setup and does automatic SSL via let's encrypt
Also got authelia wired in for SSO
For dynamic DNS i use my cloudflare domain as that seems most reliable.
@LowLevelTV is this true
Has he uploaded?
pov: pov
Yeah yeah destroy you
thanks microsoft, i didn't want to rebuild my system anyway 

you spelt it wrong, its MicroSlop
eeee
Microsoft said: "Best browser now is edge"
i'd use edge over chrome tbh
What did they measure from?
my ass its that good
What 38.6%
That's all I used it for. To download a different browser 😆
oh, its the best browser to download chrome, yea, becuase its the ONLY browser you have on a fresh install
What did firefox even do lmao
Firefox just sitting there in peace
if i had to guess, based on market share, fresh linux installs, thats the firefox metric, same for safari, fresh osx systems
windows sadly still holds the majority of systems, but thanks to the AI memory shortages, apple is the affordable option and are quickly becoming more prevalent in work as well as homes












that
