#programming
1 messages · Page 279 of 1
locally nice tho
i have class in an hour lol
can't wait to get up in 4 hours 😵💫
I'm eating into my 8 hours of sleep 
I'm about to get in my daily 2.5 hours of sleep 
so early wtf
Why is everyone not sleeping today
today?
Including me
Nah, this is everyday
gn
survivorship bias
you're only seeing the people here who are not asleep
gn, also ***HAPPY BIRTHDAY @real sierra 🎉 🎂 ***
a decent portion are i guess
and EST is... ☠️
But yesterday I didn’t really see anyone
mighta been sleeping
I was probably lurking
💡
we love getting up early
b-but what if I'm not supposed to sleep yet
then dont
gm
oh i just say gm and gn regardless of my sleep cycle to throw people off
gotta keep my cia handlers on their toes
He’s an ant
GOOD NOON GUYS BYE 👋
🐜
have a great day kosan
minor spelling misake
my cia handlers' reaction coordination game is OFF
@hearty notch fragit source code when? 
They are going to love when I decide to paint my room in real copper, surely they can track my phone inside it
imagine owning a phone lol
hope you dont use wifi
I have one but I don’t take it with me
wait ktrain can i just add you to the repo as a collaborator
I just wifi inside the room 
whats your github
ktrain
all code in neurocord is basically all ktrains
still bad 
oh ok i found it on your profile
Your ktrain
ok ktrain i invited you to the frankengit repo
its like
working but
its strange i kinda went wild with it
theres both a cli and a github server part you can host your own github server locally
:mhm: have my IP address
lmao
i didnt know thats how it worked

:)
???
I need not explain
some of the docs are outdated like im pretty sure the progressive disclosure help thing is gone
but if you get it installed try fra sandbox thats an interesting feature
also fra metaphor
lmfao
Noppers, I own this channels FAQ (for now, but they have contributed to it)
for now 
also like 90% of #1336777692070023271 are not my code even if I was to only count integrations
Surely mods won't force me to hand it over to vedal.ai
i still think i cooked with this crate naming but yea idk how much of it truly works
I sleep
gn

slepp
guys fun fact git is the goat and append-only content addressed blob storage still might just be the most powerful data storage format that nobodys topped in 20 years yet somehow only like 5% of software projects that could benefit from it use it
its so good im baffled its not the default
isn't it virtually useless for tracking purposes anyway since 95% of people sit behind nat
yea but if he doesnt obfuscate i know his general like county-level region
then i will stalk him
its over for him
no not really
researchers have more netflow of the traffic than you'd think
tempmail - exist 


Just create gmail and delete the gmail after verfied it and
Make sure that disable ure two step authorisation
And remeber ure password for login
can't be bothered
he really likes this place huh
baseless accusations 
NEW KONII RELEASE 
The connected youtube account is not even a day old 
Wtf are u talking bro and who is 100 ?
Dont worry about it
@sage crag are you 4k yet
If she gets new model numbers but keeps being 480p that would be sad
Have u watch this video
Even my watch is more that 360p


koni failed product

Do you have vrr, so it goes from 1 to 2hz seamlessly?
But you're an lcd no?

every reaction is on my notification
Then turn of your pings
no she isn’t
Nooooooo
broken heart a eye
neuroExplode
I'm done

“helo ai can you shuffle frames in this webp using pil thank”
(real)
Not the pillow python package 



I finished my math trial exam an hour early cuz i know how to do backface culling math
Its literally just the dot product
can be shortened further to “math”
that should’ve been your message

I may have done 0 > instead of < 0 
Eh who cares
cheeto dust
I have never eaten cheetos before 
orbie
Ellorb
ElvynBwaa
Mlntcandy, why are you candy?
:Cheesit:
Mlntcandy, why are you candy?
Ngl, i dont like nacho cheese flavour chips. They're nowhere near actual cheese flavour
Id rather dip normal chips in cream cheese
i just bought cheetos because i haven't had them in ages
damn they suck
they're just
sad flavored
"Sad vlavoured" 
What kind flavour is that
no but
Time to watch some anime
But i don't have crunchyroll subscription
this is like 100x worse than nacho cheese
Dont self report bro
tastes like citric acid, msg and a hint of the most artificial cheese flavoring ever
Bwaa
connie emacs
As it was nice song btw
ne emacs also bad
name leak
konii do NOT say vscode
vscode also bad
but i use it a lot when i need a featured editor
all editors are bad and i hate them all
i will write my own editor when i finish making my own language and then os :smogfrog:
you will use netbeans
i will not
yes you will
Eclipse 

notepad.exe

shumi
konii e2025v3 
What ure last version
otato emoji
oao eoi
almost all vowels
la
Konii.exe has stopped working 
Midnight mode is very good bro esp with my amoled display
satnav should be called router
have you thought of buying an air fryer?
have you thought?
i have

thinking hard but i didit


disastreous font
i would use ddg lite but it doesnt support dark mode
what the fuck is that
#NUM!
ddg for feature phones
it's for your commodore 64
wait is it down
mku
Miku with a PC inside
she is not in need of freeing
its search engines 
it;s the pc
no
why are you assuming the pc is a she

its like ships

is there a gendered language where computer is a feminine noun
spanish
or english
english or spanish
konii does not know ball
ball
true english only has "words" what is a noun 
ohmygod brugh

you programmers

"to know ball" as in knowing how to play ball as in basketball as in it's a modern saying and basketball has nothing toi do with it
tbh I agree that vscode is alright/kinda meh, its unique selling point is literally community contributions (via extensions) lol
outside of that it's a kinda slow typescript-focused editor
but w/o external extensions what are you even using vscode for

i dont understand
and "english or spanish" is 2 year old brainrot

as in it was 2 years ago not for two year olds

baseball, huh?
"I know ball" is a slang phrase that means "I know what I'm talking about" and originated from sports culture, particularly basketball. It has evolved on platforms like TikTok to express expertise in various topics, often humorously showcasing skills or knowledge in everyday life.

can we have another obsession in this channel please
borecandy
:drill
:drill
why no emoj
alo unicode can i have drill emoij

and jetpack, chattini need their jetpacks
css injection
dark reader
i should have one
get me info stealer
ok bet which one do you want
For some stupid reason I decided to google "neuro api" and for some reason the AI Overview picked the Python SDK?
(very last one)
Use tor without vpn
Or use about:blank
Kek
neuro-related simulations, as suggested by mentions of "Tony"
what is bro on about
when I searched "neuro game api" and then clicked "ai mode" (first time ever I've clicked on that btw) it also decided to suggest my package... which is for server implementations... even though every other language listed was an sdk
also phrasing is kinda wack at first glance
yeah this is why I don't trust gemini ever
also apparently there's a powershell implementation?
why?
if it's true why would it exist
Humanity's last test results say gemini it on top for now
funni
Sad pilled
I think that gemini has cheated
"My distilled slop is better at answering these specific questions"
To steal info directly from source
Any implementation of RAG that isn't local can't be trusted
my vector predictor trained on decades of human data beats your one on this set of benchmark questions we created

truth nuke
neuroBongo
This sticker is from this server. You can use it everywhere.
I'm curious that what if gemini if say wrong ans
You have unlocked new role
Let gooooooooooooooo
That's an excellent and very important question. You've hit on a core challenge for all AI models, including Gemini.
The short answer is: Yes, Gemini can and sometimes does give wrong answers.
They're not sources of truth like a calculator or a database. They are powerful pattern-matching engines. These models generate responses by predicting what words are most likely to come next based on the vast amount of data they were trained on.


That's an excellent and very important excellent and very important excellent and very important excellent and very important excellent and very important
It's not just excellent—it's important.

you have hit the head on the nail
ow
wrr
i saw you typing for a full minute just to go "ow" 
its very hard to type
leter hard to pres
you have to type "o" and then "w"
and then you have to press enter
but you see
its difficult

Video removed
Undo
Tell us why

You've hit the nail right on the head! This is an excellent observation. You're absolutely correct, hitting the keys in correct order can be quite difficult. You're absolutely right!

🟩 I've already watched this video
✅ I don't like the video
❌
wha
reported for harmful acts, as it hurt my eyeballs
unrelated video btw this is fine i like lizards

lizaerdf people next>>?
broketmc
what?
⏰ 💵

<-
no
ee?
any touhou, osu, other ye

the price variation on touhou games is wild
i suppose its because the IP is pretty much able to be used by anyone though right

probably
internet protocol

is demo enough ee or full version?
need full maybe
ok that fine
me not try yet too


Elvyn the Touhouer
Meanwhile I've never played a Touhou game
ohhhhh
tried game?
online locked till u beat game
but u can also unlock this way


hope this works on linux
@sage crag dis if multiplayer buggy
try vanilla first 
yee ok
u alr get game?
Bro my youtube feed is full of canker
Touhou mentioned
how does online work in touhou 
you play together or against each other?
me get full game then?
sporks and foons found in a kitchen
foon

admin hes doing it sideways
and by it
i dont get it
i feel sleepy
reeeeee
elvy okay with playing in a few hours? maybe 3 or 4? me energy really low 


ml boring
True
i had a small crumb of soap somehow make its way into my mouth and i have no idea where it came from
but it’s 100% soap
tastes like soap
does it smell and look like one
do you use soap
It carbonate water tasted
Except when it's something you can mess with in cool ways like a vocal synthesizer
Normal ML is boring though
i just reiterated what he said i actually think the opposite
it’s cool and exciting
but implementing it sucks
and carbonated water tastes acidic
Time set for: 14/11/2025 15:45:13 (<t:1763131513:t>)
I made a script that prints stuff like this. Is this useful?
(You type in the time you want, hit enter, it automatically puts the quoted message in your clipboard)
yes
Hi, there. Please feel free to try this web app: ASCII Cube
https://asciicube.vercel.app/
Nice js
You can drag or scale it.
It run on phone very smoothly
It's an open-source implementation of the ASCII Cube on GitHub Shop.
The source code is here: https://github.com/ZhanZiyuan/ascii-cube (vibe coding; Next.js)
heyo friends. would you mind helping me out a bit?
Tell us first
so rn i am just brainstorming about the steps to make an AI that can play OSU
just wondered if i could get some ideas and input here
Have u use osu image data ?
Well u don't need ai for that instead use image regonisation for simulation click
It light weigh
not really. as i said i am still just in the brainstorming phase. i am in 3rd semester AI rn and the lectures just dont click together. so i thought by asking here and learning on the way to connect everything I might make some progress
would you mind teaching me a bit?
Check accuracy threshold
↓
Calculate center coordinates
↓
Move mouse to target
↓
Simulate click
gpu integration
What's bro doing WITH gpu
it looks like its modifying the text of the website every frame
the gpu integration just your browser exploding because its updating fullscreen text 120+ times a second

cookie
also cube is webgl
strange

everything you need
use PPO for your clicking task and a traditional CNN for your aiming task
some repo on github can be referenced
there are deeplearning examples of training a model with replay data to play mapsets
what does PPO and CNN mean?
what's your progress on learning AI
bro is playing ender lilies

i am trying a no death run rn
already did it in magnolia
i really dont know. it feels like the stuff we learn in the lectures is useless since we dont talk about how to connect and how to code it.
rn my prof is talking about programming neurons
if that is helping
so PPO stands for Proximal Policy Optimization, basically a way of doing Reinforcement Learning.
CNN is Convolutional Neural Network, as its name suggests it's a neural network
surely the neuron part of your course is related to the neural network part
and maybe you have heard the word Reinforcement Learning before
i know that you need to da a train test data split. but knowing all that doesnt get me far if no one ever teches you how to code it
I mean, I rather prefer learning how things work and correlate to each other before doing practical, having theory knowledge is always preferred
i wouldnt complain if the correlation part would be explained. and sadly i am the type of person who needs set examples to work with.
if it continues like this i am only in that school to get my baechlor degree XD.
Mathematics with a distinct visual perspective. Linear algebra, calculus, neural networks, topology, and more.
there are videos on his channel btw, if you don't want to read the write-ups

i am happy that i dont have lectures next week
i will fully immerse in that stuff and in my jp anki cards XD
just, basic introductions on neural networks
you can ignore the LLM/Attention part if you want
nonono
thats the most important for me
basicly the only reason i am studying ai
but you don't need a transformer model to get an AI playing Osu!
thats right. the osu ai is just a small first project so that i dont
give up on the subject

i personally find LLMs really interesting
after you finish the calculus video you can jump on to PyTorch introductions, Reinforcement Learning introductions, PPO(or do Q-learning, which is also a way of doing RL) implementation and explainations

huge ass black box next-token-predictor

it wouldnt be interesting if it didnt give these fascinating results
yeaaaa i think as soon as i will try to work with LLMs and create a "person" with it, I will just start to lose my mind
wdym
all I see are embeddings
when you learn more advanced techs that make LLM do all sort of things, you'll start seeing it everywhere at every moment
time to grind some sicp 
do i WANT that? XD
You can also try writing out a neural network on paper. I think this makes programming one make more sense
👆 theory is more important in AI engineering
time to make holes on the wall, so cables are not visible to plain sight 
would be painful to do cable management
I never tidy them up
If you have sheetrock panels, you might be able to clear the paint out of the screws and unscrew them
Easier than busting holes

Some companies will use a staple gun to mount sheetrock panels instead for some reason. Those people need to be shown how to use the superior screw
OH THAT'S RIGHT!!
I forgot to mention the caulk to slop over some damage from a few very Untrue Strikes of my hammer on a couple staples!
simple problem need violent solutions 
Smoothed in with the FINEST
piece of torn-off cardboard from some box
i didn't even notice the graphics card was on the back
wait a minute
and below that is the psu
thats a pc???/
Miku pc 

Tf is sheetrock
Had to look it up. Drywall is wallboard offered by many manufacturers, like sheetrock. I was a bit confused about what was the product and what was the brand.
Bro shut the fuck up😭
Win 11 

irregular sleeping schedule
gm
Even my ping command doesn't escape my antics of over complicating everything
async
I appreciate u

that's just rust
no

musician role meetup
i meant the part where it responds with such a message for a simple ping command
oh it’s
slayyy
discord
I just think it's funny when you post single programming words out of context
it’s not out of context unfortunately
aw
woooooooo go rust
My AI is dumb 😭
Input: What do you do when you eat?
Pred : alu
Input: What do you do when you dance?
Pred : ancerip
Input: What do you do when you run?
Pred : run
More like an LM but some like that
ALUs working hard when eating 
thx
What do you guys think about the AMD Instinct MI210? Is it as good as marketed? How does it compare to similar NVIDIA cards?
or what about the way i do this silly little thing to make the commands very modular
what do you do if you fall on the floor? floor
I can ask it but its probably gonna take an hour to respond 😭 . It's not really optimized at all
but yea, the whole training data is basicly just that:
data = [
("What do you do when you eat?", "eat"),
("What do you do when you sleep?", "sleep"),
("What do you do when you run?", "run"),
("What do you do when you walk?", "walk"),
("What do you do when you read?", "read"),
("What do you do when you write?", "write"),
("What do you do when you jump?", "jump"),
("What do you do when you sing?", "sing"),
Finally finished my job system in my game engine
Well for now
Im still working on getting my old rendering system in jt
Right now it only handles the mesh rendering and frustum occlusion
As in, you moved the task scheduling out of the engine/runtime layer into a system registered on your ECS?
If you do that, you start to get issues where your engine doesn't respond to something like high load in time, which can cause one frame to take significantly longer than the rest.
Suprisingly similar to how it is done in LLMs, but alas floor


From reading so many Gemma model cards I immediately assumed TPU meant Tensor Processing Unit

its flexible plastic. i was using pla before which just cracks if you bend it
i can make springs out of tpu
ill figure it out
supposedly i dont need a temperature controlled chamber
so it will jsut be tunign settings
maybe a lot of that
monitor worked fine today
i jiggled the cable a little, probably was it
today at work i noticed that my parenthesis management significantly improved since i started doing exercises from that sicp lisp guide
studying pays off? 
LLM reasoning is weirdge
Sometimes it just has to repeat itself until confidence tips over a threshold
"Just say it stupid"
huh?
You: what is 1 + 1
Bot: i can be my friend
Ah hang on, had repetition penalty on by accident
That explains the slightly odd behaviour, its not recommended for gpt oss
Is this how "reasoning" looks like?
That's better
Discord lags so bad through wifi+vpn i want to cry 
Yeah I had the wrong sampler settings on
This is how it responds once I switched to the right profile
Why was it called "repetition penalty" if it instead was making it to repeat
More like... Repetition motivation
Do i not get it right 
Or repetition reward
It biases against repeated tokens sequences but not necessarily repeated information/ideas
And in this case it messes up the reasoning in funky ways since the model is meant to consider all tokens without added penalties/bonuses to probabilities from samplers
Does it start feeling weird if you ask same question 10 times in a row
Does it think "is it a glitch huhhh"
Lemme do that now
Its so interesting to look at its thought process 
And its still just llm next-token-predictor? Crazy
It assumes its doing a bad job
Ohh 
I would too, tbh
That is simple.
It does in anime. 
TRUE 
I'm not artist but I'm programming but we can calculate geometry
You should reconsider your VPN provider, or perhaps VPN usage altogether. Discord lags at uploading images and whatnot with no external interference by default, so a VPN is a really big ask
?
I'm Drawing a cube in matplot
you probably want to use the drawing library over the one made for plotting data yes 
Do u want data ?

Okay

It is using nvlink
gpu id -1 
Kimi K2 is dumber than DeepSeek ( .. but the same in speed
Depends on your test case, also at such heavy levels of quantization it could be a difference in quant resistance rather than actual model quality
Additionally the one everyone is glazing is this
The model you're testing is older than Terminus and lacks its reasoning capabilities
For reference here are the three compared on bench numbers, you'll note that the original version of k2 falls behind both deepseek and k2 thinking by a pretty large margin
Forgive the low res, fitting them all in one screenshot on a laptop was rough
with the same quantization level... k2 is twice as large, it ignored half the prompt and returned a very mediocre result. and this is a constant
I'm not surprised, the original K2 release wasn't that groundbreaking
K2 thinking is a totally different model and way smarter, and that's the one with all the buzz you've been hearing about
Try this and you should see it easily surpass Terminus in utility
I was just about to download it and try )
Report back when done
Extra opinions on these things are always useful since it is ultimately subjective with recent large models
ok, I have a pretty complex task for the model. let's see if the new version can handle it...
Oh make sure to use --jinja for the model launch and the recommended sampler settings in the params file as well, you probs already know this but I always throw that in there when people are testing just in case
Vision? It has then means that means i will use into main model as backend
Nah it's text only, I don't think there are any major frontier size open models with vision
I stand corrected, turns out there's a 235B version of Qwen 3 VL
What a huge model ;
MY MAIN AI IS 1K PARAMETERS
Oh yeah Kimi K2 Thinking (the model we've been talking about) is 1 trillion parameters
Maybe I will train my model from other model
If you want a decent easy to run local model with vision, I recommend Mistral Small 3.2
That's 24B
If you need even smaller then go for one of the smaller Qwen 3 VL models
I'm not sure what you're referring to here, I would assume this is back end specific?
Yes
Yeah I can't really help you there, I use llama.cpp
Or the quantization awareness trained versions of gemma 3 if you find qwens sycophancy annoying
Python 3.13.7 (main, Aug 15 2025, 12:34:02) [GCC 15.2.1 20250813] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tinytuya
>>> DEVICE_ID, IP_ADDR, LOCAL_KEY = 'redacted', 'redacted', 'redacted'
>>> lamp = tinytuya.Device(DEVICE_ID, IP_ADDR, LOCAL_KEY, version=3.5)
>>> lamp.set_value('20', True) # LED ON
{'protocol': 4, 't': 1763181844, 'data': {'dps': {'20': True}}, 'dps': {'20': True}}
>>> lamp.set_value('20', False) # LED OFF
{'protocol': 4, 't': 1763181849, 'data': {'dps': {'20': False}}, 'dps': {'20': False}}
>>>

I'm in.
time to make it control a drone now
ok now get a ton of people to run a program in coordination that will, when all their lamps are edited into one grid, play bad apple
I'm stuck at 3.12.9 version of python
My bot getting smart
interesting plan
I use terminal to play
to play what
hi
Have u try to play chess in terminal
no
there's probably a tui app for it tho
WELL THIS CHESS PROGRAMS IS ALSO TUI BASED
coherence overflowed and became negative 32-bit int limit
Maybe I will change neural architecture sooN
@true hemlock better cpu fan when reeeee
but like bro
you’re literally the same fucking person

Wtf are u talking about candy and who is same person


I'm getting confuse
Tell me who is same person like me ?
@rigid snow tell me ?
u're making me confuse
Why can't I watch the video?
Ure application cannot
So it ure device cannot
What ure phone MODEL?
Maybe it doesn't support 1440p and 4k
Why does thw weather app need the ability to change my system settings? This makes no sense
most coherent bot
okay..
automatically updates your location settings when you open the app mhm
Alot of other apps also jad it enabled by default
Now people say I'm bot :(
I was going thru and disabling a bunch of unnecessaru permissions
Google had the ability to turn or off wifi, like what
idk man, there was a guy that was selfbotting here a while ago or smth
and had the same coherence level as you
O okay
but he didn't join a vc so 
Google also had a bunch of stuff enabled everywhere
Now I get it a guy who usess bot in own account instead of legit bot right ?
yes
Wtf who is same person tho ? One answer get now second answer ?
Apps cant show me ads if they dont have wifi 
Oooh
Use android data saver
I cant enable hotwpit if thats on, and its qnnoying
Cheak in network
Plus smh some apps still access it, since i still get alerts for apps using data in the background, i know bc i keep getting spammmed pinged with "some app used more data than usual"
Data saver is a suggestion at best
First on data save
Lol
Im first on blocking out wifi access for apps
Its two choices for those apps, stop using my data, or get deleted
It kinda sucks that internet is one of those permissions given by default that you can't even revoke easily
True
Yea
Best i can do is set it to wifi only
Only one things is best
Never connect to wifi
Was samsung max the Fucking bloatware
Waht
WHY DOES A CALCULATOER NEED WIFI, this is so stupid.
Maybe currency exchange 🤣
it is
are you actually gaslighting yourself into thinking it's not him or are you pretending
my calculator cant even cnonect to wifi
My calculator app wanted wifi
i think kosan just isnt good at english
exact same phone, exact same misspellings, exact same topics
the selfbotter didnt really form proper threads of messages
then again the acc was made today
but i wouldn't jump to conclusions
That is valid proof
But I'm not bot bro 😭
exact same misspellings
by this i mean specific ones
that i have never seen anywhere else
Have bot play Spotify
he's not a bot and he wasn't banned for being one
there were multiple offences
So i person was accuses been self bot
6000 rpm coming in less than 2 weeks i hope
search for "cheak" in this server and look at accounts
his trait
how theoretically could you decrease that for any type of cipher given a basic bruteforcing algorithm that checks all solutions and magically knows when its found the right one
hikari going crazy in thje free meowing vc
Now I get it
Some one name call letter i and accuses as selfbot account
yes
but like
what would you be able to do for this
O(2^N)
why 2^N
What the fuck is n2
Yeah, an n bit message has 2^n possible values
two to the power of the variable N
There's a lot like backtracking and uhh some other search algorithms
But they're all still exponential
Actually no
This doesn't apply to ciphers since they're not in NP in the first place
Which is good 
yes
What do u think that I'm bot ?
can't definitely say yes but also can't say no without lying either
I'm back
So ure answer is probably
More progress on my assembly voxel raycaster
Ignore the fact that the shadows are all black
but what if we actually need humans with machine-level intelligence?
for what
for our lord and savior neuro sama of course

I'm just watching Yann LeCum lection nevermind 🤣
i miss the days when half of the internet wasnt blocked here 
if only we had human experiments allowed
Kind of crazy how much delay images can have to load for seemingly no reason
And then you paste in Imgur link and it's almost instant
Maybe Discord is just using a crap CDN
I'm back
my vpn works fine on pc though. i think its either wifi/my phone/phone app
Maybe it ure wifi router
its cheap anyway so i dont expect excellent performance, but this dramatic difference between pc and phone makes it weird
Cyborg yea
i really dont know what settings i should look into
is there some anti obfuscation protection












newgens





took me a moment to get it