#programming
1 messages ยท Page 318 of 1

if miyagi sing this song, it almost in character
why caption that
i think it admirable 
what if audience is near-deaf but want to enjoy music anyway
until either you hear or feel it depending on how deaf you are
just realized headphones don't work for completely deaf people 
Provided to YouTube by Nettwerk
The Red Means I Love You ยท Madds Buckley
The Red Means I Love You
โ Madds Buckley LLC under exclusive license to Nettwerk Music Group Inc.
Released on: 2021-02-24
Producer, Engineer, Mastering Engineer: Josh Sebek
Producer: Charles Ritz
Producer: Madeline Buckley
Mixer: Megan Chase
Music Publisher: ASCAP
M...



https://music.youtube.com/watch?v=KSfTEtrM1AU
more edgy music 
Provided to YouTube by ASIAN MAN RECORDS
I Love You Like An Alcoholic ยท The Taxpayers
"God, Forgive These Bastards" Songs From The Forgotten Life of Henry Turner
โ 2012 ASIAN MAN RECORDS
Released on: 2012-06-19
Main Artist: The Taxpayers
Composer: The Taxpayers
Music Publisher: Songtrust(BMI)
Auto-generated by YouTube.
What new GPT wrapper should I make?
Also, should I ask a ChatGPT wrapper this question, instead of asking #programming ?
yes all slop related questions are best contained in the slop bubble
Or should I just make another ChatGPT wrapper to ask the ChatGPT wrapper to ask the question?
Oh
example of non-slop question:
- how do i make cupcakes?
example of slop question:
- i made chatgpt sentient, how can i publish my research?
please self identify and appropriately ask the correct channel

Presenting my sophomore album SUNK COST FALLACY, out now via Pilot.
Download/Stream: https://foxstevenson.ffm.to/sunk-cost-fallacy
--
Tracklist:
01 - YAS!
02 - Curtain Call
03 - Give Me Some Space
04 - Exile Is A Habit
05 - Road To Nothing
06 - Tryhard
07 - That Choice
08 - Interlude Canyon
09 - One Horse Town
10 - Memories
11 - What Are You ...
it album 
too lazy to look up me build ingame inserter speed calculator instead 

Provided to YouTube by The Orchard Enterprises
i wanna be your girlfriend ยท girl in red ยท Marie Ulven Ringheim
chapter 1
โ 2018 world in red under exclusive license to AWAL Recordings Ltd
Released on: 2018-09-14
Producer: girl in red
Music Publisher: UNIVERSAL - POLYGRAM INTER OBO UNIVERSAL MUSIC PUBLISHING LI
Auto-generated by YouTube.
good album
(i already listen)

look like 0.8/s
album
most of my current music taste not something you'd find on streaming platforms 
that what soundcloud for 
true!
pyanodon make you build this for 0.4 belts of copper plates 
at least ore consumption only 2 belts
me forgot have construction bots
not played in long time
run out of copper for formaldehyde
not run out but
not enough
looks like i wasted a day trying to figure out what broke clangd just for it to turn out that clang simply doesn't support cuda 13, good to know i guess 
wdym clangd doesnt support cuda 13 
how does that even work
does cuda ship their own compiler? 
nvcc is based on clang
i'm not sure what exactly is the problem but there is a problem which in turn breaks clangd too
based on 
nvidia should ship clangd too then
do you know about the hobbit shire? (russian band) 
most of the useful code is in llvm i'd imagine since they have their nvptx backend, just need a wrapper to handle the separate compilation of gpu code etc conveniently 
i not know about russian music anything except for rappers (studio gangsters)
i listened to them a lot in school
obvious but most songs lotr inspired 
phantasy rock ye 0 chance
@sage crag useless fact that band exist
now you know

finished interstellar
they used a lot of bass in the mastering
had to actually lower the subwoofer from +10dB
It would seem our cynicism was misplaced https://www.pcgamer.com/games/rpg/swen-vincke-says-the-price-of-ram-and-ssds-means-larian-will-be-doing-lots-of-optimisation-in-divinitys-early-access-that-we-didnt-necessarily-want-to-do-at-that-point-in-time/
the best part of the ram shortage is that devs will have to give a shit again
and kill off Unreal Engine

Neuro is controlling it
let them play 
they'll suck but it'd be funny
OM
+NF 
One single circle clicked 
I guess the .npy stays for now, I'm actually working on neurons somehow.
if I don't die in the dev process, then I actually want to do a demo. But that requires not dying 
One mistake I'm not making this time is lack of base class.
Why am I even spending so much time on type hints? My main editor doesn't use them.
i want to bind my right windows key to f21 on linux and for some reason key_f21 is not the same as f21.
it doesn't need to be f21 specifically. any key not on most keyboards will do.
input remapper does not like it
classic
I ended up doing something more stupid in the past
wrote a xkb for my own usecase
let me sanity check rq, I want to see if it works
minecraft was a bad game to test with
I just opted to go nucular and modifiy a keyboard layout for what I was looking for
i don't know what to do with that :P but i might do the same
but if you know of any other key that is unused
xkb has a lack of docs, man I spent so much time getting a file to exist
sounds rough
let me get something 90% of the way to yours
well the reason i want a special key is i'm creating a game and want a secret key to use
// the name of `nameoflayout` is the filename
xkb_symbols "basic" {
// american, use your base
include "us"
include "nameoflayout(begone_key)"
};
//
xkb_symbols "begone_key" {
replace key key <RWIN> { [ F21 ] };
// clear the Super modifier part of Super keys
// there is a 50% chance this works :shrug:
clear Mod4
// readd Left Super
modifier_map Mod4 { <LWIN> };
};
This should be close enough, minus the exact keycodes
i don't know where this goes. i'm new to linux
That is a good question, I was just able to source it without needing to actually figure out the correct location
oh lol
fricking Nix, I can't use the traditonal location
well i might use some other key that works for me
~/.xkb/symbols/my_layout is what the AI says, it seems correct 
do you have a "menu" key on your keyboard?
either that or yeet it /usr/share/X11/xkb/symbols as where the other ones live
My issue is that I've done my testing on a laptop 
let me see if I can hunt down the defenition of the super keys so I can at least complete the example
I really need to ask @fast pagoda for the chatgpt character prompt again, my version is clearly high
what does this last part do?
How much detail do you want, I have two versions of the explaination
what is mod4, why do i need to map a modifier?
Names are dealt with, nice
Mod4 is the Window modifier (Some call it Super over on Linux), we are clearing it and readding the left one.
do i need to do that?
If you skip the deletion of the modifier, then you can still do Win + key with your remapped F21 key
i see...
Think of the name of the key and the modifier as seperate
like there is a left and right shift, you can bind each key in games if it supports it; but both act as a shift key
I hope that makes sense
yea like on windows, win + L would lock the screen and it would work with winr too and if i don't delete the last part, when i press "f21" + L it would do the same still.
That is entirely correct
ok how do i use it? reboot? select it in settings?
We'll have to figure it out
let's try this one first. So that'd just be placing the file, naming it the name without file extension and reboot to refresh, we hope it shows up in keyboard options.
just make sure the name you pick matches the name inside the file itself
-# because I am using vesktop now... I can acutally use a .XCompose file for emotes 
i don't see it. its 5am i gotta go to sleep. thanks for trying to help!
All good, you'll have to make the file path yourself tho.
have a good rest
It still doesn't fricking work 
It literally works in the terminal but not in discord 
I messaged you Mr iggles

how is the arg arging

can we remove the alignment requirements on vector instructions pls 

horendously
sure
ill allow it
I spent like 3 hours trying to implement overlapping buckets in my data structure just to realise there is no overlap with 256-bit loads due to alignment

i dont get what you mean by overlap?
Think of a sliding window
yesn't

but why?
it's like doing list[1:3] into list[2:4], the data is already there so just hand it over 
Oh wait I misunderstood
This is done for optimisation is all that matters
It helps but needs a lot of overlap

Ideally each index would be the start of a new one
i know i can jsut change the offset for reading the next vert in vulkan
but i doubt you're usign vulkan

No vulkan just (good) old cuda c++
cuda c++ good 
Unfortunately it's the least shitty option

Anyways, big fan of Blackwell 256-bit loads 
Just make it not require 32B alignment pls

unfortunately
no gamba 
wdym no gamba
the money was never gambaed for stocks
I came back here because of the only time I've seen AI point out a very useful feature I'm 100% using in the NN repo
Well I got the Discord bot api working for my model so I can now cry as I watch it have an aneurysm when I pull it into a call with friends
Apparently numpy dtypes are composable
I was going to dedicate an entire axis to this, but if I can just do it in the dtype then sure
time to create u128 rq
u128 is now real
still limited to byte alignment on the smaller end, but you win some and you lose others
not a real u128, but it works for exactly my case of "I need a larger than 64 bit field"
brother, why the frick not?
if you cant make your cpu faster, may as well give you 16 of them to do things in parallel.
that's how the computer can run minecraft bedrock edition so smoothly, because eventho its the inferior version, it being multi-threaded proves that no program should ever be written in java


let me set the names to a range, do it
I want to be able to have types like uint4x32
I can generate the types, that is not the issue. I just don't want to convert numbers into strings for fetching.
considering how numpy's array notation works I can see why, but I just don't want to.
-# sooner or later I'm just going to be writing staight C in python, mark my words
the what repo
nn has a repo??
overall, this is a good change for me. It's going to be more painful sure, but it'll be fun to make.
I guess I will actually have to have rock solid broadcasting rules now
I'll also need some very special types, oh boy this is going to be a massive scope increase from the previous version.
I guess it is a v3, so it had to be insane.
i got a j*b woo
good
Ewww you traded having a life for the money you need to have a life /s
One of my friends (recass) made this amazingly fast, transparent and auditable AI architecture: https://github.com/recassity/cassity-model
And she said, she thinks its scales via CPU power. Cause it does barely any math, it just a recursive system. Inspired by how the human brain neuron works. I wanna see someone experienced in this field, review and rate the architecture for her.

your friend look like this?
Wdym my friend?
I'm confused.
sorry i meant like this
Wa-? She's a person.
What did you even mean
Oh

On the topic of this. Anyone's gonna test this out?
its gibberish
What's gibberish, the repository?
Now that I've read it. At first glance, it kinda is...
its also written in pure python, no gpu acceleration, no simd
Theres no need to be

Theres barely any math operations
Its lile a recursive linking tree
yes i saw the linked list, im not sure what linked lists have to do with machine learning
Have you like, tried python -m cassity on it, and said hi?
Idk either, its not used anywhere
https://github.com/recassity/cassity-model/blob/main/cassity/src/cassity.py
This is the main file that has most of the main logic
it reads as clearly vibe coded, i dont know what to say
Could try codespaces, no? Disposable pc
Its okay!
not that my time is particularly valuable
It isnt?
Did u mean, I will get kicked?
likes looking at people's so called "magical AI tech"
the fricks given in here are scarce nowadays I see

Does this count as a markov chain
only so many times you can see the same crayon scribble on the wall before it stops being endearing and becomes infuriating
ancient technology

oh I do agree. Took longer than I expected tbh
dismantling such code will never not be fun though 
im tempted to give it a code review
might kill me though
No idea
ok
Was this sentence AI generated?
The - thingy gave it away
Its not on normal keyboards
no, i wrote it to look ai generated
i copied an em-dash off ddg
i can also just type long dash anyway: ใผ
How does it work from your analysis?
just a hunch 
On windows u can press Win+; to access a bunch of symbols includin this em dash
Iirc there's also some combination with alt key but I forgor which one it was, never use those
alt+0150 and alt+0151 for en and em dash respectively iirc
Oh I almost forgor why I even opened this channel
Question, if an artificial intelligence uses diffusion model to generate audio content that was not in the dataset it was trained on, that's gen ai, right?
iso linux keyboards you cna press Ctrl+Shift+U and then type 2014, and press enter
maybe works on ansi keyboards as well
yes
true, rare, me just lurk here usually. i only really chat in #gaming and nn rn
From my knowledge it is but some person online is tryin to convince me it's not for some reason
Maybe I'm missin smth but I looked it up a lot and I just dont understand what could be wrong
gen-ai refers to using any kind of model to generate content based on its training data
even if it doesnt spit out something that is in its dataset exactly, its still genai
The AI in question is SynthV AI voicebanks that use deep learnin to create more realistic voices btw
It uses Diffusion Probabilistic Models, accordin to the Dreamtonics website
Dreamtonics releases the Synthesizer V Studio 1.8.0b1 update. As a major and comprehensive update, 1.8.0b1 is being rolled out as a beta first. The update introduces a major evolution of the synthesis engine - high speed Diffusion Probabilistic Models (DPM) based singing synthesis technology, along with a number of exciting features and improvem...
interesting
when we refer to diffusers we are typically referring to stable diffusion because its a colloquialism, but diffusers themselves are just a model architecture
it is generative still, by nature, though
whether or not that's bad should be down to the dataset (rights) and what it's used for
Yea I just mentioned somewhere online that generative ai is not inherently bad and gave SynthV as an example, since it uses voices of people who gave the right to do so, plus it still requires a lot of effort to create smth. And some person is trying to tell me it's not gen ai. But like, technology used is gen ai.
gen ai isnt "bad" (in the sense that its pretty impressive and has so much effort put in from very smart people), what's bad is consequences of gen ai like people being deskilled or pushed out of jobs, internet being filled with generated garbage, etc
what also can be bad is how genai development proceeds, it's entirely possible all the effort spent on transformers/diffusion/associated datacenter construction could be spent more productively, but its just spent in most profitable way 
Diffusion can in fact, be used for singing voices
i didnt say otherwise

that's a given, diffusion is a model architecture and not a specific purpose of model
There's also UTAU which is very cool
what the hell is that
my ears have been pained ty
im going to have nightmares
(mostly by the spiky bits or whatever idk how to describe it)
have to process utau vocals a lot
compress compress compress
since the banks are made with unprocessed recordings and then the effect is even more amplified by how it slices recordings up

Some of the i sounds of Teto EN are indeed abit, weird
Yeah that one's like completely unprocessed
Not to mention Teto English is inherently just like that for whatever reason
-# What I sent is a raw mixdown
Teto's Japanese voice bank is much smoother though, likely because Japanese is a lot simpler phonetically
Who knows, but it definitely works without much processing
Apparently "nuh uh words are subjective"
What is this, people spreading lies on the internet? Surely that doesn't happen
ASM is 3 times faster than C++ and 2 times faster than Rust.
They said they studied this in uni
I imagine the explanation would be smth like "generative ai is ai that makes something new but you can just call whatever you want as such it doesn't really matter, end of the lecture" 
Weird stupidity
That is, of course, if they even had actually studied that
Perhaps an artist
"It's just a copy from multiple sources"
can this dumb ass discussion stay on twitter please
genai is not an insult it's a term that means the ai generates something it's that simple
Yep it's literally just a term, don't argue about it
NeuroSynth could be considered generative but it's not hated because it's not the same kind of generative as SUNO but instead the same kind as SynthV
yea i agree
but how do i explain smth to a person like this 
Pianos don't use machine learning, and that's why
bro's onto nothing ๐ฅ
imagine they argued back with some bs theory about human thought are the same as machine learning
That'd mean the person is clearly an idiot and nothing to care about

didnt you know
vocal folds are generative ai
What is generative ai? An ai that generates something?
I love ai but i just donโt want ai that make vid or drawing i want ai that can fill the damn excel spreadsheet for me accurately 
probably gonna make one myself
generative AI I believe is just deeplearning models that are capable of generating contents as a black box

idk
there are machine learning branches trying to simulate neurons tho, they're not entirely wrong, surely 
generative ai = ai that generate output statistically likely to occur following specific input
piano = not ai
me think argue with this person waste of time 
me consider gen-ai as ml model that designed to create synthesised content based on training data 
scroll up
animal maybe but human eh the odds of achieving that is lesser then the odds of nuclear war happening we human barely understand our own brain and thoughts enough to replicate it.

oh DPM 
i have things to say that i shouldnt 
nuclear war tomorrow 
DiffSinger in SynthV ๐ฅ
"shouldnt" in what sense 

it has standard voicebanks that work by just combinin the samples recorded by a real person, but most voicebanks use ai trained on such samples instead
like
teto or gumi dont even have standard ones for example



I see
jam
boom
seem easier to do for jp becuz not as complex as en

oh and same for Vocaloid6 btw it also uses ai
tho iirc it has very little voices so far, but like Miku should appear soon? it was supposed to be in 2025 but the year is about to end so prolly not

ye ml sounds so much better is why
worth dealing with the "all ai is bad" crowd
all ai is bad (piano)
and you're one of the more competent people here
rename channel to #baking #incompentence
do ai even exist
but AI consumes too much water
*incompetences

becuz
help preventing sea rise 
true
would be better if enub
non artificial intelligence also consumes water.. kill all intelligence then
mlp technically not fit as "ai"
that
nub
league players would be spared (no shower gang)

Machine learning way better term
wrr it because field started with thing like chess ai
for what
I don't call NeuroSynth AI, I just say machine learning vocal synthesis
marketing terminology anyway 
they donโt mean the same thing
Well classic ANN models
baking with your CRT electron beams? 
make chess no longer fun tbh

why would i do that
hav oven
chess seem more competitive than ever though
online chess help to but me think engine big part

you use oven
so you are human, with built-in oven
correct?
correct
maybe that happen in case of caibi ye 

oven is an enclosed compartment, usually part of a cooker, for cooking and heating food. My stomach is an enclosed compartment, capable of cooking and heating food 
kill everyone NOTED
nobody spared
which food
which food exactly
are you cooking at 37 celcius
yuh
is stomach acid hot? neuroThink
low heat
no?
obviously

so then how will it cook
it wont
acid can be hot with certain reactions
idk tho
surely there will be exothermic reactions taking place 
eat
oh wait they actually does that

crystal sodium hydroxide

hey if wiliamosman can drink motor oil and survive would it be fun to try
complex words nowaying
its not even close to as dangerous
what does that possibly mean
get those neutralizations in my stomach!!!!
depend on the molar concentration of the acid
ouh

that's the oven overheating and burns itself through
it all make sense

idk chemistry tho only chemistry ik is from nilered most of it the explosion and toxic stuff

apparently avg vol stomach acid 0.02 to 0.1 L, average pH 1.5 to 3.5, so worst case you have 0.1L 1.5pH, 0.00316 mol of NaOH to neutralise entire stomach acid
which like, less than 1g probably
after that it start melting you

melting as in what type

turn into slime
new tortura method dropped
there's a reason why even if you drink tiny amount of sodium hydroxide drain cleaner (like 10% conc by volume), they have to pump your stomach contents out
its also incredibly unpleasant way to go apparently
can HDPE contain such puddle
reinvent previous

need to try on sam altman

ceo of closedai
ye i dont think sodium hydroxide can corrode hdpe
so no matter what volume
should be fine

bit spooky
aside from neuro i have been enjoying crackhead engineering and chemistry much lately
bases are a lot more scary for organic compounds than acids
chemist here
i dont know the context
but indeed
and that guy who have 100 battery
very unlikely
Heisenberg taught me Chemistry
usually you need very agressive methods of digestion to degrade materials such as plastics
not important 
context: melt person
please use enjoy chemicals with safety
HF
with alkali specifically
you could try concentrated HF
you need special protections tho for HF
its the one they used on Breaking Bad
that scene is a tad overexagerrated tho

but there are research studies in which they digest small animals, like mice, in HF
this is the high seas, what is safety 
i did napkin math on sodium hydroxide and stomach acid 
fuming nitric acid
very unlikely
animal studies are highly ethical and not scary 
thats true btw
ive done research with animals
ethics committee are a pain

humans caring on research-use animals when they can't even handle themselves well
maybe bad search "how much of XYZ acid to dissolve human"

animals have no way of consenting to being experimented on
so they best be careful

tbh is hypocritical cause usa really like japanese research on life subjects post ww2 and its not even animal
I heard they consented, lgtm 
na just go aperture science way
black mesa my beloved
you need to test medicines and treatments on someone
its not like they simply spawn
ethics both make science advancement slower but also important.
now tho, theres a lot of research being done with AI to improve the process as a whole tho
that being said: medicines are still tested in humans
Peak
it's called the clinical phase of a medicine
can we make ai robot then inject them instead
curious
im well aware of this whole dynamic
medical ML model actual useful 
then, if, say, only n% show bad results to the medicine, they allow it to the open public
do we need consents for testing protists
one of the most noble uses of computer
sometimes they WILL collect medicines from the market, as the population react badly to it (sample sizing issues lmao)
that with ai applications on making the disabled life easier imagine what good robo with awareness could bring to the blind
Truly cool stuff
skill issue gene diversity issue 
wrrr
Silly
:neuroTrue:
instead human use it to make shitpost on tiktok
Yup. It's hard to get a representative sample size for a country size population in a clinical context.
Now, imagine that you have (in a big, well develop, doctorate level) study in a good day 50 animals
Sometimes not even that - it depends a lot on many factors.

If any of you are concerned tho, research is (in any decent university) conducted under rigorous ethics rules, which are displayed in any crediable journal (usually at the bottom of the paper). Also, there are several "rules" (they not rules, but rather some aspects) that are taken into consideration when designing an animal experiment.
For example, you use mice for many medicine development tests for depression, anxiety and so on
Some utilise pigs.
(For other stuff tho)
animal cruelty is the norm outside science people just not like to think about
meat industry dairy industry
terrifying
ping
they can be ethical prove : country in sea region but choose not to
True that's a pretty SMOL FPS
is it supposed to run with a single core or something
it just simple triangle 
It is working as intended
It's usually CPU or RAM limited
how the hell did caibi get 10K on an intel iris
what kind of drug is that thing using to be that fast
My usage sometimes
iGPUs can actually get great scores on this one since less latency
becuz u no need gpu
are there mac ver of these
Yeah software render
there is no way that my chip is faster than an rtx 3060 LMAO
Yeah triangle is a CPU benchmark more than GPU
& u need to figure out how optimize
And I guess RAM probably
customize compile flag
RAM speed and latency and stuff is important for triangle benchmark

I wonder if my triangle score was more limited by CPU clock or RAM
Since I've since increased my CPU clock to 5.2GHz
quack score so unbeatable
True nobody's ever breating that
that assumes quack will forever have bleeding edge hardware 
Help me please
I got hyprland and i don't have any window for streaming in discord
How to fix it?
ye rebooting won't hurt
can also try using the web version of discord
the desktop client is weird sometimes because Electron
smoll
its indeed mostly cpu dependant afaik
cuz every gpu in existance that is newer than 20 years can easily do a single rasrter per frame
in nanoseconds
since software rendering is on the cpu itself there is way less latency, same ting with igpu
the only way dgpu get decent fps in this is by making them run behind the cpu without making the cpu wait
so thats why you can see some good scores on actual gpus
new guy workin at local fry place.
ordered burger. receive burger, no sauce, no bread, no vegetables, no anything besides the meat.
i have received the jsut the meat

was it cloudburst who ran the triangle on apple silicon
what was the score
oh vani
6k only 
ye
ye needs to use MoltenVK
but still
would still expect over 10k at least for an M4 Pro, yeah
even with some extra overhead
it might be some issues witht he manual core assignment and stuff
actually ye probably
some platforms use 0-1, 2-3, 3-4, ... for threads on the smae core
others do 0-n, 1-n+1, 2-n+2
i cant math, dont look at how many edits this has
the M4 Pro doesn't even do SMT AFAIK

if so then it should be fine
unless it has mixed core types
iir cit should be using core 0 and 1
which benchmarked best on my pc
If you try to use standard thread affinity methods (common on Linux/Windows), you will find that macOS does not support pthread_setaffinity_np for pinning a thread to a specific hardware index. Instead, the "index" or "type" of core your code uses is determined by the priority you assign to your thread
that might have been an issue actually
wait, im an idiot
#ifdef _WIN32
...
#elif __linux__
...
#endif
it doesnt even try to set thread priority on macos 
if i want to actually test my code on macos i might need to buy a 2ndhand mac mini or something
if you develop stuff that targets apple devices having an apple device is generally a good idea yes

Sadly yes
im not targetting specifically, i jsut want it to work there too
Which is why for the game dev work i used to do we only supported Windows and Linux
idk if im willing to implement metal tho
Finally
not targeting ios 
We supported Android
Ios would never allow our games
bye
I over eeped 
and new subathon mashup
im watching shutter island with my sister in a bit tho so ill have to dip
how is bro running on 4090s and still lagging? 
you say this as if you haven't bashed VRC yourself
looks at world
neuro does look like she has 6 digit poly count tbf
Very Poor
100K seems likely to me
we love hair collision

i do not remember reacting to this 

auto-react or dementia
did you happen to maybe double tap on mobile
it's really easy to trigger accidentally
ah

so its dementia
then idk
got it
time to figure out how the world affects lag

Welp thats at least working... Honestly Id have to say that it is both encouraging and disheartening watching vedal still struggling with Neuro doing the same things Ive been running into
that assume he stop get new hardware
well the 4090 is neuro's computer and not his right
4090 in his as well
Depends on resource scheduling. It may be able to handle everything but process scheduling is bad when dealing with multiple compute types
I hit that wall already, I bet he's aware of this though
Most likely guess is his VTT plus vr plus nvec is thrashing the GPU IO, not overloading the core/memory
Colored lighting on

so what you're saying is that he needs more computers with more GPUs clearly
Yes and no, problem is he can offload VTT but it's honestly best running locally and serving the text externally. It would explain why neuro is having trouble understanding him sometimes, when VTT is overloaded it drops info
It's not super resource heavy locally but it is sensitive to utilization spikes (like gaming)
so it might've been fine in testing because he wasn't streaming so no nvenc?
I need to stop piping hardware deals to Quack's dms then 
Exactly, something like that could 100% push the setup beyond its capacity. Each task running simultaneously has to compete and while usually pretty good at handling it, nvidia's scheduler kinda breaks down if it's handed a bunch of tasks that aren't at least partially resource aware
could switch to x264 if that's the case then, although that would require restarting the stream
Would explain why he mentioned he might have an idea but he would have to restart the stream
@sick owl wtf I keep getting impressed by nano banana pro
i took it to add an extra 'sub goal' from the screenshot I gave it with 4K setting, it actually upscaled the screenshot and added the sub goal flawlessly
Crazy isn't it
I'm continually impressed by its capabilities
wtf thats nuts
it even got the number coloring in correct order
and it managed to shift them all so that it is still centered
also post that in #livestream-chat as it's pretty funny ngl
obviously I did LULE #livestream-chat message
|| it got ignored basically
||
ooh report button exists now, I can't wait to not use it
lawlessness in this chat
@ nixos users
I am not here
is there anything specific I need to think about if I want to move an old crappy convertable laptop to nixos
its a foldable touchscreen thing
Nothing besides pray you don't need to compile drivers (I'm looking at you nvidia)
Standard linux issues exist ovbs
That's a problem you need to run into imo, not likely to happen
ye just need to look at drivers for the various devices it has I think
can check https://github.com/NixOS/nixos-hardware for if someone already looked at it

drivers should be fine, I think I can always remote build
sadly not
its an acer aspire 3
pre-usbc ๐
check the arch wiki
and then map whatever additions might be necessary to nixos options
but not very
it has a usb c port slot on the motherboard
but no port
@cinder trellis if NDI is an issue, and causses lag; you might seek to use srt;
have your Neuro obs record to srt your pc as sender, and on your stream pc have an srt source. Important is that the sender side records to srt and not using streaming to. That will make the latency near 2 second or less.
so for sender (unde record to url) : srt://your-server-ip:port
on receiver side as vlc source : srt://your-ip:port?mode=listener
make sure the listener is active before you start recording.

the advice seems genuine at least
We train LLMs to accept LLM neural activations as inputs and answer arbitrary questions about them in natural language. These Activation Oracles generalize far beyond their training distribution, for example uncovering misalignment or secret knowledge introduced via fine-tuning. Activation Oracles can be improved simply by scaling training data quantity and diversity.
Chain of Thought in a diffrent flavor
portofolio
bonus money 
where was that github repo?
@jagged turtle getting replaced 
.
I hate whatever this scuff is
Since some time after the intro of the subathon my stream started lagging unbearably and nothing makes it stop doing that, not a reboot even
I have literally no idea if that has anything to do with my PC crashing
alrighty desktop site works, the mobile just doesn't
I pray that Vedal restarting the stream fixes it because nothing I'm throwing at it works
vedal definitely not a web dev
that's like the first thing you learn
mobile-first and responsiveness
Probably AI-generated
oh yeah but it's not like the ai would implicitly decide to make it work on mobile
im not able to buy fractional stocks on my bank's app 
ktrain's repo at least works on mobile
i wanted to buy the same stuff she's buying for funs
as it should be
I hate this lag AaAaAaAa
I hope it's not a hardware or driver fault
I hope I didn't corrupt my system when it crashed and I force reset it
Got my embedding model running on the A2, it's now fast enough for it to be a viable option for me to be able to split memory into active, short term and long term and reduce session context down enough to rearrange resources on the main server. Should boost performance significantly along with the existing optimizations I did today
Why am I spending so much dev effort on fricking typehints?
AI devs don't even use typehints, what is the point in this?
I don't think it did AaaAAAAaaaaAaAAaAAaaaa
windows update 
NixOS mentioned
Anything above 480p feels unbearably laggy still and I don't know whyyyyyyyygh
nixos is mentioned almost daily lmao
let the jank begin
Oh I just tuned in
But like what is the issue? It literally just appeared and like after my PC crashed or something
The subathon intro worked flawlessly
I don't even know how
But like there's never been any issues
ALSO the issue effects Streamlink
Which has no GUI
And no encoding or decoding
I dislike this currently but I don't know how I can do anything else atm 
honeslty idk if twitch hardware acceleration even works on my pc. its at 0%
I don't even really care about hardware accel I have plenty of CPU I just want my 1080p stream back
And I want my Streamlink to not lose frames
Yes
This feels a ton like some sort of Twitch connection issue but I don't know why it appeared specifically after the subathon intro
is your GPU still being used or have you switched over to igpu
something could've done a driver update and become unhappy
I was about to yap about the CUDA VA-API backend but apparently we already talked about this ages ago 
should work
#programming message
My iGPU never worked as an offload on Linux so I bought a second dedicated one instead as an offload for the VRAM heavy tasks instead, so everything happens on one GPU
dec
Well doesn't look like that's being used then
I have plenty of CPU power though and it's never been an issue before
This feels so much like a degradation in Twitch connection quality and I don't know why it appeared specifically when it did
I did not want scuff like this on subathon day, this is so annoying
I'm having such a bad day
Maybe I just need to sleep and hope it fixes itself
Hopefully it's just some stupidly timed maintenance of some kind and it goes away tomorrow
tableslam
on the other hand, free CSS 
is his stuff almost entirely css
nah, the entire app seems vibecoded from other chatters
it does look somewhat vibecoded
Mobile just is non functionial via my (clueless) testing
it's probably not designed by an ai i'll tell you that much
the test u64x4 type works. That is good
what the fuck
Anyone have a recommendation for a stand that can hold the ighty GIRTH of a P73 Thinkpad?
Every laptop stnad I can find is 100% not for a 17 inch
Yeah something like that but
This laptop is like 20 pounds
I bought a brick I was planning to take to work and clas with me
Lol wth. My laptop weighs 5.8 pounds and already has way too much power and huge heatsinks.
anyone into theoretical physics?
Depends
suspenseful
On?
On what areas ur referring to
Alot of different theories out there
I have a theoretical degree in phsyics.
Julian Barbour, GR, Quantum, Jakobi?
Yeah this thin is 7.5-10 lb
THICC
thats wild
Yeah the building blocks of our universe are fascinating
It makes my 15.4 MacBook Pro look like a netbook.
We've still yet to find the graviton. Or take GR into quantum.

i really started disliking the ThinkPad plastic after having a machine made out of metal ngl
Im not sure how I feel on the graviton idea
Sure
you can keep doing physics here
I'll keep going
This is the server's nerd-mall
im doing geometric algebra
wave functions yet?
ye, but not in-depth
its fun
Im heading to bed soon but honestly ive mostly been following string theories concept of gravity for a while
I love it. I love my Macbook too. But I might sell it. Unless I turn it into a LinuxBook Pro
. No clue what OS to use though, it might be Ubuntu and I disable the AMD GPU for batteyr reasons.
the wave functions are more used as a way to and end for the transformation matrices we're doing
string is poop imo
Did you try NixOS? ๐
The graviton seems worse though because it screws alot of general and quantum mechanics
Nope not yet. I've mostly just sticked to the top 10 on DistroWatch
Well if you're ever feeling adventorous and you want something super reliable and indestructable go check it out ๐
from what i heard gravity is a property of the fabric of the universe bendign or something like that
i have no clue

-# you also need to configure a lot to actually get those benifits
true but its not that hard to get a basic config working
once you have it you can reproduce it any time on any machine
and it literally cannot break
It can't save you from yourself, but it is pretty good otherwise
make it a flake and store it in git. then you can install your entire machine from github
The thing with gravity is if its really tied with time as a dimension then I dont know if we should expect a single particle responsible for it
i mean with my current setup i literally cannot destroy my system. its impossible.
well, you see. Jakobi and Barbour came up with a 'timeless' universe. The theory is pretty solid.
I'm framing my work off them
I mean it depends on what people are calling time
exactly.
I view time as just moving one directly 4th dimensionally
I view time as a memory of change within a Local Time Dilation zone i call LTDZ
4th D would be where gravity comes from
imo
But when you go QM, its hard to say
I mean gravity is usually associated with the curative of spacetime
system not just programs/configs but also state
if your system state can be trivially replicated after nuking it you're indeed quite lucky
Schrodinger is too hard to tamper with and not break it.
its trying to play osu




me not use hyprland
