programming

[flint dome]

harvard cs50 is free

[glad path]

I dont mind

[brisk laurel]

Do you have a recommendation?

[azure lynx]

was already typing "googling for stuff is also (unironically) a helpful cybersecurity skill. "
sorry ;]

[glad path]

w3schools is one i personally use

[brisk laurel]

[flint dome]

or freecodecamp on youtube

[true hemlock]

though don't rely too much on the courses. sometimes you'll have to build a project to sharpen your skill

[glad path]

its more of a reference

[flint dome]

cs50 is good since it gives assignments to do alongside as well

[true hemlock]

the courses are there to help with your basics

[brisk laurel]

Im currently doing job employability skill courses rn so year

[flint dome]

it should also have a cs50 python version

[glad path]

don't be afraid to google a LOT btw

[brisk laurel]

[true hemlock]

and don't be reliant on clankers

[keen hatch]

it all depends on the tasks...

[flint dome]

https://cs50.harvard.edu/x/ @brisk laurel

[true hemlock]

just don't be a vibe coder lol

[azure lynx]

pretty much anytime you come across something new (especially things you don't understand immediately or that aren't clear), google the main words

[true hemlock]

what are you doing with it

[flint dome]

has a basic of everything you're looking for

alternatively this is good too https://www.youtube.com/c/Freecodecamp

[azure lynx]

once you've learned one programming language it's much easier to learn more

[brisk laurel]

My professor taught me prompt engineering even though its not related to our course cause why not considered he is already certified

[keen hatch]

I'm currently fine-tuning a Ministral-3-14B on it (~100gb vram) ...

[flint dome]

ooh das big

[glad path]

ai is pretty useful to learn stuff btw

if you just ask it to walk you through and explain and not write the code for you

you can also paste ur code and ask for help

[true hemlock]

that's fair i suppose

[true hemlock]

i still couldn't believe it that "prompt engineering" is a thing

[brisk laurel]

Im on 1 percent

[keen hatch]

For other tasks, I have a different system...

[brisk laurel]

He has given me templates for prompt engineering :3

[glad path]

wont that last 4 hours

[true hemlock]

oh that's pretty fair

[flint dome]

whats your major anyways then

[brisk laurel]

I didn't use my battery saver this time cause i need something to upload

[brisk laurel]

IT

[flint dome]

ah

[brisk laurel]

Dont ask my minor

[flint dome]

what year rn

[brisk laurel]

Last?

[true hemlock]

i was just saying machine learning prompting is literally the most basic part of the whole field, like anyone could do it

[brisk laurel]

This year

[flint dome]

ok

[true hemlock]

i couldn't believe that they called it "engineering"

[flint dome]

final year? why they no teach you programming till now fahhh

[brisk laurel]

Okay my phone us shutting down bye chat

[brisk laurel]

[true hemlock]

anyway

#1310060238099058799 message

[flint dome]

charge it

[brisk laurel]

I know and its only 1 year

[brisk laurel]

I know this

[true hemlock]

obligatory for programming starters

[brisk laurel]

I cant im in uni and i dont have my charger with me

[azure lynx]

if you can reliably explain something to a person, then you could do it. I don't think that means "anyone could do it"

[flint dome]

i think this prompt engineering stuff might be helpful

finally might help normal people how to google

[flint dome]

borrow a cable

[true hemlock]

i still wouldn't consider it "engineering"

[flint dome]

yes

[true hemlock]

felt more like an insult to the engineering field in general

[azure lynx]

technically... yeah. no. yeah.

[flint dome]

would not go so far

[brisk laurel]

It is helpful
Considering my prompts are usually like " you are a professional interviewer with 20 years of experience on IT and interviewing. Ask me questions about my field and general questions. What i got wrong and what i need to improve. Do not make a mistake. Think hard about this."

[flint dome]

we were told in smaller classes how search engines look shit up in our comp classes
so it was easier for us
now they maybe dont tell or just are plain lazy

[brisk laurel]

What about social engineering

[amber fractal]

prompt enginnering is a very high praise for basic skills you lack

[true hemlock]

im saving that line

[azure lynx]

social engineering is a valid skill. it makes some in the engineering field uncomfortable because it involves interacting with people though.

[flint dome]

that is a sort of thing most programmers usually throw at it and know by default, but for a normal person i would appreciate taking a short course on it

half the time we dont have to throw those lines in either because the ai already picks up on intent from the language we use (atleast for me it does)

[amber fractal]

you got me at the last second

[brisk laurel]

Idk any of this im cooked😭

[amber fractal]

Chat, before you slander it

[brisk laurel]

I know

[flint dome]

dont worry about it for now this is mainly just a baseline

[amber fractal]

how many are you confident in answering?

[brisk laurel]

0

[flint dome]

i could miss out on maybe 1 or 2 but rest i think i could do

[azure lynx]

how many make sense as questions? (like you understand what they're asking?)

[true hemlock]

i dunno i don't work with general full stack.

if the questions are more HPC and hardware related i'd be more confident lmao

[brisk laurel]

Also
Igglyn —> iggly —> igglies
Is an interesting development of your name

[brisk laurel]

Im a beginner no

[true hemlock]

nah he's Intanaani !

[true hemlock]

though im definitely familiar with every single question

[brisk laurel]

I read that as itani

[amber fractal]

[flint dome]

give truth table of IMPLY :trollface:

[azure lynx]

i understand all of them, but can only answer 6 or so comfortably. i'd use google in real life for the rest if they came up.

[amber fractal]

igglies has always been the identifier, iggly is my normal and igglyn is an alternate

[brisk laurel]

Also you probably don't remember me but we used to be friends and you helped me alot somewhere in late 2023-late 2024

[azure lynx]

but also, i've had lots of experience.

[brisk laurel]

I see

[amber fractal]

What is actually suprising is how many sites didn't already have igglyn as an identifier

[brisk laurel]

Its a unique name

[amber fractal]

discord was one of those

[azure lynx]

maybe you made an account before but forgor?

[true hemlock]

you're not the first one i know to throw out the n from your name's last letter

[amber fractal]

I've been as iggly for years

I decided as I was getting more active to switch it up and add the n

seems like to this day, a lot of people think it was the reverss

iggly was a shorting of igglybuff. Still my favorite orb from pokemom

[true hemlock]

oh

welp the opposite with those 2

[amber fractal]

@primal dragon

alts not allowed in neurocord

[final tinsel]

i just googled my username and all the results are just my accounts lol

theres nothing else

[true hemlock]

alts are allowed i think

[amber fractal]

your right actually

alts are not banned

I just have no reason to use it here

I was probably thinking before the rules rewrite

[true hemlock]

or you were thinking of the other server

[amber fractal]

nah alts aren't banned there. I have it on there

[empty rivet]

there was a small boom of courses that teached prompt engineering in my country and some people made a lot of money with it

[jagged turtle]

social engineering is a thing tho

[loud salmon]

Has anyone heard about backdoor cyber attack?

[jagged turtle]

???

[loud salmon]

The attackers, I call them pirates, they attack from the "back security system" rather than hacking from front

When you hack a computer, you use it like a prop to attach the main C and C severs and they get played by downloading a file contans the viruse

[obsidian mantle]

What is this "back security system"

[loud salmon]

A backdoor attack involves a cybercriminal creating a hidden, unauthorized entry point into a system to bypass normal security, allowing them stealthy, persistent access to steal data, install malware, or take control, often remaining undetected for long periods by exploiting software flaws, using malware, or phishing

[loud salmon]

to short terms the first trap when is set if the hacker falls for it gg you secure the system

but if he pass the tricks he indeed experiences or next lev of hacking

this is level one of many other

[lament igloo]

all roads lead to image generation inside cloudflare workers

i never thought i needed it until now

[loud salmon]

explain

[lament igloo]

sometime ago, i was experimenting with dynamic image links that generate an image for discord embeds

https://neuro.nya.pub/neurotimer

this is one example

[loud salmon]

ay no way you fall for

this

[lament igloo]

this is also an example https://354-five.vercel.app/api/image?text=Elly_(TH)

[lament igloo]

wdym

elaborate pls

[loud salmon]

no need

[lament igloo]

?

anyways i dont wanna use vercel

cf workers is where its at

🙏

[gentle fox]

Tech talk!

[loud salmon]

Whatever you feel comfortable, you can use

[gentle fox]

Welp, back to NN.

[loud salmon]

i use what's better in performances to get the best experience

[loud salmon]

man.. what most i hate is the shortcuts in words

[lament igloo]

im not spending a single dime

is it free

[fleet jungle]

This sounds cool

Imma try it

[loud salmon]

you have apoint

back to mining crypto

[true hemlock]

social engineering is a real engineering/study

[jagged turtle]

???

fall for what

[fleet jungle]

social engineering is where hackers reside

strange bedfellows and all that

[faint sandal]

[unkempt citrus]

Social engineering is where I, an engineer, make plans for a date

[fleet jungle]

epic

[olive sable]

https://tenor.com/view/azumanga-daioh-gif-2910224911933025207

Goodmorning

[opaque wharf]

Giving me himotou umaru chan vibe

[olive sable]

[patent shard]

I thought expirement may have something to do with expiring, but I actually don't think it's a word

[lament igloo]

its a misspelling

[glass jetty]

yeah firewall and dns servers just a joke

[maiden geyser]

"Has anyone heard about backdoor cyber attack?"
starts explaining
asks nothing else
disappears

[opaque wharf]

Chad

[glass jetty]

can some one plz do it for me everyday

[glass jetty]

do you want me to explain linux interrupts?

[loud salmon]

idk i dont feel well maybe i will take a break and hoping one day i die

[loud salmon]

if you say so, my brain is not fungtioning well i feel like im melting inside

[loud salmon]

i dont know what im say33ing

sry

[opaque wharf]

[glass jetty]

you type, not talk

[loud salmon]

i wish if it was a dumb act from me but its not im really feeling not gud

[maiden geyser]

fruit so low it touches bedrock

[glass jetty]

fruit?

[maiden geyser]

agree, at that depth it's probably called a root vegetable

[glass jetty]

what?

for the last 2 month or so: each time i try to understand how interrupts invoked in first place i go over 3 stages:

1. ok, i remember interrupts work like that and have traits that and that, and also interrupt context cant be interrupted but after handled in context it go and invoke something with schedule, so interrupt context is only head
2. who tf enable interrupt line back then? is it linux kernel or hardware? how interrupt invoked in first place by cpu? how it saves registers and where? let me search
3. fuck there nothing new i can find, there literally no good source of info, id try later

[maiden geyser]

bond name's the james

[flint dome]

job had one bro

[obsidian mantle]

Is it possible to make a window stay focused

On windows

[maiden geyser]

wintoys?

[jagged turtle]

there's a powertoys utility called alwaysontop

in terms of actually focusing the window, no

idt

[obsidian mantle]

okay ill check that

Oh wait always on top not enough

Whatever i'll see

[noble zodiac]

The bootloader and kernel both enable interrupt handling. Software and hardware events can trigger an interrupt. The cpu executes the corresponding handler according to the IDT. The registers are saved on the stack by a special wrapper

[sage crag]

[true hemlock]

size comparison

[obsidian mantle]

What is that fin on the right

Titan is so good looking

Is it loud because 1 fan?

Or its some super high end silent fan

It looks unusual

[true hemlock]

fish

[true hemlock]

it is loud. its a blower style fan

centrifugal fan

people don't like it for some reason :(

i think blower style fan has really good aesthetics

[obsidian mantle]

Is he okay

[olive sable]

Exam went

One of the randomised questions was "how do twitch streamers make money"

[nocturne olive]

Hem

What kinda exam is that?

[opaque wharf]

Should've answer neuro style: By your mom

[rare bramble]

lil overcooked, still very edible

[nocturne olive]

Tasty fish and chips

[obsidian mantle]

It looks like cpu but 10 times bigger

[unkempt citrus]

Parasocial relationships

[olive sable]

Game history

[olive sable]

Its multiple choise

Answer was "all of the above"

[amber fractal]

What were even the options?

[olive sable]

Donations, advertisements, spinsors

And another one but i forgot

[rigid snow]

de-dieing the cpu

way better cooling performance, never goes above ambient like this

[opaque wharf]

Merch?

[rigid snow]

for streamers/content creators in general it’s 95% sponsorships and brand deals

in similar fashion for music artists it’s not what you’d imagine either, it’s playing shows that makes you money

everything else is pennies unless you’re taylor swift

[rigid snow]

youtube pays relatively well though

[olive sable]

that might have been the other option

i forgor

[fleet jungle]

Have you seen the Cerebras processors?

[ivory plinth]

sand undertale

[jagged turtle]

hopefully this is vedal-proof ux

[amber fractal]

I give it 50/50 chance

res = np.equal.reduce(arr2, axis=1)
No it isn't real, thankfully

now I get to work on the correct solution for that particular bit of code.

[jagged turtle]

I mean I'm sure vedal-proof ux basically is just "send that shit to neuro"

[real sierra]

are you kidding me

let me get this straight

if your class has a constructor with arguments

you just

cant use it as a component

because why would anyone ever need a constructor

whatever man im so over it

[opaque wharf]

Godot

[gritty dust]

@olive sable my final RST is in 2 hours and my exams are like 3 days away im so screwed mannn

[real sierra]

thats fine just send another SYN ACK

[gritty dust]

I need a syn ack with my brain mannn

i gotta get ready for class, ill see you, holy im failing this

[amber fractal]

[real sierra]

WAIT

wrong channel

sorry

u got this

[frozen hollow]

I never had an issue with this - isn’t this kind of an antipattern anyway

Ike it very quickly becomes hard to reason with the system type of antipattern

Unity is a legacy behemoth tho idk I hope godot replaces it for what it does.

[real sierra]

everything is an antipattern in unity

there doesn't seem to be a pattern that actually works

the unity way would probably be to attach this to a gameobject, then prefab the gameobject, then clone it every time and instantiate the property

but that just seems like more work than creating the gameobject dynamically

so why would i

it would not be a hard thing to add so i think they should just add it

[frozen hollow]

I think the performance and traceability of adding components that way sucked I don’t remember but there’s a good reason to do it with prefabs - something batch-related

[real sierra]

thats great but its not a high-performance application

i think the option could safely be provided with the appropriate caveats

but it simply isnt

[frozen hollow]

All I can say is it’s easy to make the performance 10,000x worse from putting the wrong conceptual system into Unity

[real sierra]

oh well

unity's fault

[frozen hollow]

I just work with what I have

[amber fractal]

that's good and all until you have nothing

[real sierra]

if unity didn't make things a pain in the ass this wouldnt be an issue in the first place

i dont even want these objects

id rather these class instances just be in an array somewhere

but alas, if you want unity functions you must use a monobehaviour, and if you want a monobehaviour you must be a component

and even if there's workarounds for that, the networking library im using further screws me by subclassing monobehaviour

i have a square peg and unity will either present me a square hole or i will make one

[frozen hollow]

Idk. At the absolute base level I wouldn’t try to work around anything

[amber fractal]

if your using the object sorting bucket in full, the analogy would be: "Either unity presents a square hole or I'm ripping off the lid"

[real sierra]

an engine that does not make the developer's life easier is a bad engine

godot for the next rewrite

[amber fractal]

*this rewrite

[real sierra]

*next

[amber fractal]

you'll fall eventually

the next rewrite is now

[amber fractal]

Glad to hear a differing mindset, I very much disagree with never breaking stuff open.

[olive sable]

im tired bwaa bwaa bwaa

[ivory plinth]

[real sierra]

bweeeeeeeeeeeeeeee

[lament igloo]

funny techbro youtuber replied

[idle dune]

Wrong chat

[nocturne olive]

Unity existing is the best Godot ad

[frozen hollow]

I think this is a very unfair read on what I wrote

[amber fractal]

Is that all you pinged me for? Just to say I'm wrong in what I said and say nothing to what fashion.

[sage crag]

https://cdn.discordapp.com/attachments/1061113840852533339/1452612577145262183/3DNeuroCutenessAmbushWithBeautifulScenery.gif

boring

make an interesting conversation

[olive sable]

hi konii

[frozen hollow]

No I wanted to know why you got “never” out of what I wrote because I didn’t write that

[sage crag]

hello camuel manuel danuel

[frozen hollow]

I also now don’t want to continue that off the tone though.

[wary mauve]

Is it better to put conversation history snippet in the system prompt or just pipe it through as user & assistant messages?

just a lil curious

[olive sable]

camuel manual danuel?

[wary mauve]

https://cdn.discordapp.com/attachments/701598894974435380/1452996062892523665/Screenshot_2025-12-23_at_12-07-03_Instagram.gif

[olive sable]

[sage crag]

https://cdn.discordapp.com/attachments/1061113840852533339/1452612577145262183/3DNeuroCutenessAmbushWithBeautifulScenery.gif

[olive sable]

i had a think while cycling

if we use a DLP projector

then a frame consisting of multiple bits per pixel would be smeared

and i think that wouldn't actually be a bad thing

there would be less seperation between the different layers

for our purposes we need a lightcrafter EVM, model 4500 or higher

the lower end models dont have pattern sequence mode

if anyone can find a lightcrafter 6500EVM for less than 1K, ping me

[frozen hollow]

You really wanna make anime real

[olive sable]

its not just about the anime

its about proving that project ava is bad

koolmees

i scared them off

[acoustic finch]

wel

[olive sable]

[acoustic finch]

i was gonna say not trying to defend project ava but if you have to spend 1k

you kinda destroy your own point

[olive sable]

i have to spend 1K because mine will not be a flat lcd panel

its not about the money

[frozen hollow]

What’s actually the problem with the razer one

[olive sable]

its 2D

[acoustic finch]

yes i get that i read your discussion yesterday about the cylinder

it uses grok

[frozen hollow]

I feel like it being grok-backed (?) is more significant to me lol

[acoustic finch]

i was gonna say

[olive sable]

i couldnt care less about grok, i would have just modded it anyways to take hdmi

[acoustic finch]

honestly if its not to expensive and you can jailbreak it it might be fun

[frozen hollow]

Is it gonna crash out if I play wolfenstein

[acoustic finch]

i think it would call you every slur in the book

[olive sable]

project ava is a flat lcd panel

its worthless

and they're gonna have the audacity to ask 400 bucks for it

[acoustic finch]

i mean its literally just spyware to steal your data i wouldnt expect anything else

[olive sable]

i want to make a real volumetric dispaly

not this cheap fakery

[opaque wharf]

Neuro Lantern!

[acoustic finch]

the fact that its just a 2D screen makes me feel like you should be able to sue for false marketing

[olive sable]

nah, they never claimed it was 3D

people just see techd emo and go with their own story

[acoustic finch]

i feel like that was very heavily implied in there marketing tho

[olive sable]

Brought to life as an animated 5.5" 3D hologram
i guess

hologram: a three-dimensional image formed by the interference of light beams from a laser or other coherent light source.

anyone wanne sue the shit out of razer with me?

[opaque wharf]

ENC28J60

uhh, wrong place

[olive sable]

ethernet controller with SPI interface

[opaque wharf]

Yep

Apparently my highschool self bought about 10 of em

[olive sable]

my highschool bought a "Virtual Production Stage"

that shit with them led walls

[opaque wharf]

That is an awesome school

Also, what I meant is that I purchase 10 of those stuff when I was in highschool

Found it during my great component sorting arc

[olive sable]

[frozen hollow]

Idk if they meant a 3d character on a flat screen

Razer make a lot of bad stuff idk they’re pretty random

[olive sable]

i mean, you cant really call it a hologram when its a flat image

[opaque wharf]

Now a lot of them lives in their tiny apartment

[frozen hollow]

It’s prolly gonna end up like that Covid mask thing they did

[olive sable]

do you maybe happen to have a lightcrafter 6500 EVM in there?

[opaque wharf]

You would not believe this....

But I don't have one

[frozen hollow]

[olive sable]

[acoustic finch]

as far as i am aware it doesnt realy matter right they framed it as a 3d hologram not a flat screen thats enough to be like dishonest marketing or some shi

[olive sable]

ye probably

i odnt have the money to take them to court tho

[acoustic finch]

that they technicaly meant something else and they hid that in the fine print wont and up mattering

[olive sable]

i need that money to buy a lightcrafter

[opaque wharf]

You are in uni right? Ask them nicely via email of your friends that is in EE Major

They usually provide sample for free or cheap for student

[acoustic finch]

just get neuro to be your lawyer

nothing can go wrong

[olive sable]

i do gamedev tho

[opaque wharf]

That's why I told you to use your friends in Electronic Engineering major to lend their email lol

But I guess it also depends on your uni email address. My uni used to have our major as the subdomain, so anyone can identify what major is the sender

So the email format is [name]@[major].student.[uni].ac.id

I got me.student because I am Mechatronics Engineering

[olive sable]

it seems that they only provide them cheaply to uni labs

not individuals

i cant really take one home

[opaque wharf]

This is where your social skill is tested

[olive sable]

[glass jetty]

my condolences...

[real sierra]

I was searching around on Stack overflow and

I noticed that having any special characters in the full path that leads to Unity project folder will crash Unity WebGL build at the end every time.

professional game engine

[acoustic finch]

last living stack overflow visitor

[glass jetty]

no one care about utf

use base ascii

[real sierra]

I care about utf

[glass jetty]

stack overflow old answers are soooo useful

[real sierra]

I'm also not sure what they mean by special characters

hopefully that doesn't include any ascii

[glass jetty]

wait some time until new tech will not be included on stack overflow

[acoustic finch]

oh i know which is why its so sad its dying

[glass jetty]

and you will end up with new stack overflow, cuz ai is shit

or there will be actually useful ai

[opaque wharf]

Oh it will include ascii somehow

[real sierra]

[opaque wharf]

See?

[real sierra]

it's spaces

[real sierra]

defend it again

[glass jetty]

nah, unity is c# so it is shit by default, i just talk about ascii vs utf in general

we talk eng/rus, and we dont care about anything utf related

[nocturne olive]

Unity existing is the best Godot ad

[glass jetty]

even half of our front end use std::string instead of qstring

[frozen hollow]

This new stack overflow has been marked as a duplicate and therefore you cannot post

[acoustic finch]

from what i know stackoverflow was already dying before gpt because of it being a kinda toxic place

[frozen hollow]

Absolutely

[acoustic finch]

ai just sped it up 100X

[wary mauve]

toxic af

[frozen hollow]

It was a nightmare

[velvet crest]

You have unlocked new role

[glass jetty]

who and which role

[acoustic finch]

yeah anybody asking questions got called a dumbass

[frozen hollow]

Me probably but idk which

[opaque sigil]

Embed perms

[opaque wharf]

We should make our own stack overflow

With Neuro and Evil

[glass jetty]

it called discord... oh

[opaque wharf]

(Instead of blackjack and you know)

[glass jetty]

make here forum for programmers

lets goo

[real sierra]

even if the unity build pipelines doesn't want to support utf

[little pendant]

SO had its moment in the sun. But yeah, been fading for years now

[real sierra]

it should detect that and error at the start of build in a controlled way

not randomly at the end

and even if it doesn't feel like doing that

[nocturne olive]

Simply use Godot as clearly Unity can't build a competent engine

[olive sable]

thats kinda... what this is

[real sierra]

it should at least support ASCII?

[acoustic finch]

its going to be just as toxic only response is going to be [filtered]

[nocturne olive]

Me when the random command at the end of execution they run in CMD doesn't have the paths quoted:

[glass jetty]

yeah basically unity

its ok when you get linking errors at the end of 1 hour compilation
but it isnt if it is not c++

[frozen hollow]

P sure i had those

If you mean posting images

[nocturne olive]

That's my best guess as to what's happening, I don't see another reason for it to not support spaces

[real sierra]

last time I tried to build, two platform builds failed and one build succeeded
after some investigation I concluded that one of the failed builds actually mostly worked, and the successful build had completely failed

[glass jetty]

btw, any one know do rust check any linking before the end? prob not but idk, seems like wanted feature

[olive sable]

anyone want to sponsor my volumetric display
only a small sum of $250 and i will put ads on it

[frozen hollow]

https://tenor.com/view/trade-offer-dungeons-and-fakultet-gif-15123941056434615026

[olive sable]

im listening

[little pendant]

Often has to do with delimiter handling, esp depending on variable expansion, etc

[frozen hollow]

Who are the deepest pocket weebs

[little pendant]

Just avoid spaces as a general rule

[olive sable]

variable expansion

https://tenor.com/view/satoru-gojo-domain-expansion-muryoo-kuusho-six-eyes-jujutsu-kaisen-gif-2102848127558680877

[glass jetty]

yeah, lets make path look like regex

[olive sable]

i thought i saw one for $500,but they were selling the software

[opaque sigil]

What a steal

[sage crag]

https://cdn.discordapp.com/attachments/1061113840852533339/1452612577145262183/3DNeuroCutenessAmbushWithBeautifulScenery.gif

annoying

boring

[olive sable]

according to gemini, i have "hit a very specific gap in the market"

in other words, it called me poor

[acoustic finch]

how far ai has come

i cant wait til we have them walk around in robots so that they can taunt you by waving gpus in your face

[fickle rain]

Disable the robot and scavenge the GPUs out of it

[acoustic finch]

then your mandatory neuralink deep fries your brain

[little pendant]

I like the false confidence one can get vetting responses from one ai with another. There's a term for that, but can't recall it.

Or the great idea that it encourages you to turn into a 15 part plan, then you have it evaluated for holes by the same engine in the guise of a shark tank role, at which it eviscerates the very plan it helped you construct as an expert.

[frozen hollow]

Confederacy of dunces

[real sierra]

hi programmers

[shut sand]

no

[real sierra]

I have a question

WHAT

[dark wing]

No

[stray dragon]

you don't have a question

try somewhere else

[frozen hollow]

a
t

[stray dragon]

yes one of those is me

[fluid hinge]

ask in #announcements

[stray dragon]

hi hello

[olive sable]

i have come to make an announcement, #programming is a bitch-ass motherfucker. 🗣️

they said no when i wanted to ask a question

thats right

[fluid hinge]

you mean NN, GN AND programming

[olive sable]

so im making a callout-post in #announcements

#programming you got a small ram size

its the size of my windows vista laptop, except even smaller

[acoustic finch]

honeslty should just write a script to at any moment of the day auto respond no

just for the love of the game

[glass jetty]

i can give you raw script for other task

it can find where to place text and send

[acoustic finch]

there you go

no more questions in programming allowed

[glass jetty]

Worst JS developer: "it basically do this"

Worst JS developer: "you just need to wait on socket"

Worst JS developer: "or wait for new elements"

Worst JS developer: "it do ui, not socket work"

[acoustic finch]

js and typescript make me wanna cry anyway

[glass jetty]

Worst JS developer: "nah, there no use"

Worst JS developer: "just do shit until it work"

[acoustic finch]

to be fair at my school we do basically purely web development and its either php laravel or nextjs

and nextjs is like 10 times easier to use

but it is js

[little pendant]

Oh what was that called...the sony portable that ran vista. Was used as an entry keypad in one ep of 24. N1000? Doesn't sound right.

[olive sable]

ps vita?

idk

i dont remember any sony device running windows vista

[little pendant]

UX180P

[olive sable]

huh

interesting

[little pendant]

Yeah had abysmal battery life but it was also really cool

[idle dune]

I'm curious, anyone here run a LLM on a Rasbery Pi with a HAT 2+? is it worth buying for a little cheap project machine?

[acoustic finch]

how big a model are you planning on?

[idle dune]

Well I know its a lightweight system so probably only Mistral 7b?

Something smaller maybe

[acoustic finch]

i think you probably have to go lower

[idle dune]

[acoustic finch]

i mean you could i think but it will struggle

[idle dune]

So probably like Mistral 3b or something similar

[acoustic finch]

yeah that would work quantisized ofc tho

[opaque wharf]

Why is BLE assigned number doesn't make any sense

Like, what do you mean heading field is optional

[idle dune]

I just don't even know what I would do with it. Like as a project.

[rough bloom]

IIRC it has 8GB of onboard memory, so a 7B model would probably still be fine?
the issue with going below that is that models become really stupid

could run a tiny VLM like Moondream mabe

[nocturne olive]

Yeah if you have 8GB mem you should be able to fit 7B

[idle dune]

You can* get 16gb Rasberry Pi 5's. But they are very popular for retroconsole and handheld guys so they tend to be sold out alot.

[nocturne olive]

Go Q4 to be safe and it'll fit plenty well

[acoustic finch]

its gonna be pretty slow tho

[rough bloom]

ye, but this is about memory on the HAT

[idle dune]

Ohhh

yeah I don't think its alot

[rough bloom]

Pi system ram is super slow

I don't actually know how fast the memory on the HAT is, but hopefully faster

[acoustic finch]

should be its specifically for ai

[idle dune]

So basically from what I read its a Hailo-10H module on a custom board and can do 26 TOPs.

So yeah it'll be very slow

[acoustic finch]

i mean right now im loading a 7b model on my gpu with 8b of vram but i am not sure how wel that translates

but if i go much higher i need cpu ofloading right so

[rough bloom]

ye these edge TPUs have historically not been very useful
pretty much only used for computer vision tasks AFAIK

[nocturne olive]

TOPS don't really matter, bandwidth does

Inference anyway

[rough bloom]

just too weak for anything else

[rough bloom]

they will if they're low enough

[nocturne olive]

Well maybe

[idle dune]

1.20 to 8 tokens per second depending on model accoding to some youtube videos

[nocturne olive]

But ususally the bandwidth is so low that it doesn't matter

[nocturne olive]

Not the 190 I get on a 30B on my 3090, but not the most terrible I've ever heard of

[acoustic finch]

i mean you can do the 16gb Rasberry Pi 5 right and then ofload to cpu to run bigger models if you dont mind waiting for 20 mins on a response

[keen hatch]

is important for prompt processing

[visual pawn]

I've run GPT-Neo on a raspberry Pi 4 years ago. it uhh, worked I guess

[idle dune]

I feel like grabbing a used workstation would crush it and be price comparable because a 16gb Pi with a Hat+ 2 is 500 bucks CAD

[visual pawn]

seems like they're pitching this thing specifically for LLMs

and VLMs

[idle dune]

Yeah I am

[visual pawn]

https://www.raspberrypi.com/products/ai-hat-plus-2/

With 8GB of dedicated on‑board RAM, the AI HAT+ 2 is ideal for running large language models (LLMs) and vision-language models (VLMs) locally and securely

[acoustic finch]

i have a mini pc at home i got for 250 with 16gb ram i could check how wel that does not that i have high hope

[olive sable]

im looking through the ebay listings of a company that does DLP stuff, and its all projector tech and then suddendly a Nintendo wii

[rough bloom]

ye stupid
haven't even released info about memory bandwidth

better than without HAT probably but still very bad

[visual pawn]

but it will deliver "reliable, low-latency AI processing at the edge and without a network connection"

[idle dune]

Yeah that feels like a lie

[rough bloom]

it probably does
just not LLMs at a reasonable speed

[acoustic finch]

there has to be some benchmarks on what it can run right if this is how they are marketing it

[visual pawn]

they have a bunch of sample models

[frozen hollow]

That feels quite useful for some ML microservice type things like idk just nvidia parakeet or embedding models

[visual pawn]

https://github.com/hailo-ai/hailo-apps

yeah I think this does have some use case but

I just dunno about pitching it as something for cutting edge VLMs/LLMs

[acoustic finch]

Llama 3 (8B) INT4 10-12 T/s ~5.2GB

that seems surprisingly good if accurate

[rough bloom]

int4

[idle dune]

Whats INT4, I'm really new to LLMs.

[acoustic finch]

weights in the neural net are long floating points

to save memory you can shave some of

[frozen hollow]

Definitely not cutting edge - practically speaking an embedded system

[acoustic finch]

so you can take a model with 16fp and reduce it to 8 or 4 which reduces memory without having to much impact on quality

[visual pawn]

looking at this, it's tasks that are actually really useful for me for work, except I don't need to care about optimizing space

but if I did

but like, if you had an endoscope and you wanted some embedded depth estimation

this could be good

[rough bloom]

except it depends
with smaller models the drop in accuracy at 4-bit already becomes significant (depending on how good the quantization is and what is being quantized)

[idle dune]

Oh got it, so your just trimming fat basically

[proven merlin]

(QAT ftw)

[visual pawn]

optimizing memory while sacrificing quality to different extents

[frozen hollow]

Yep yep there’s a lot of low-pressure but moderate RAM tasks I have running that don’t need to actually be on my pc you could shove onto it

[acoustic finch]

that could be i have acces to a lot of gpus so i dont run many small models

[acoustic finch]

yes its generally worth it

[sage crag]

https://cdn.discordapp.com/attachments/1061113840852533339/1452612577145262183/3DNeuroCutenessAmbushWithBeautifulScenery.gif

[proven merlin]

Q8 is no brainer and Q6 is nice. Quantization Aware Training on Q4/Int4 is what i prefer but no models

[idle dune]

What do you guys use your local LLMs for? I'm trying to find projects but alot of the ones I keep finding at like "Make a custom chatbot" or "get VLM to work accurately" which does sound fun.

[idle dune]

https://cdn.discordapp.com/attachments/1061113840852533339/1456314196655472843/SPOILER_NeuroCuteJogInHerRoom.gif

[acoustic finch]

i build a mcp server for my internship that allows the model to execute readonly querys on a db

and now im looking at finetuning models

[proven merlin]

Mostly giving me research keywords if im unsure what i actually am looking for, video search, template generation and yea.

[acoustic finch]

i do think mcp servers are a fun place to start with ai

[frozen hollow]

Grab something and run with it

[nocturne olive]

I don't use them for much of anything, except Xoda's school project needs one so I'm helping her with that

[visual pawn]

I've been trying to find a use case for one in surgical navigation, but I haven't personally found any meaningful success where a non-LLM wouldn't have been more useful yet, but I do know of some papers that use a VLA as part of a pipeline for automating surgical tool control

[idle dune]

[visual pawn]

one of the draws for having one in a pipeline is being able to claim that your automation is more transparent and thus helps with trust, but I also just straight up found more success with canned messages

[acoustic finch]

i think using mcp servers could be a good starting point it gives you a lot of options

[proven merlin]

Is there a difference between mcp and tool calling?

[idle dune]

What is a MCP server?

[acoustic finch]

no mcp is just the protocol

[acoustic finch]

its a way for a llm to interact with the outside world

[visual pawn]

one application would be using a local LLM to make an assistant

alexa+ but local

[acoustic finch]

the llm generates a tool call in a certain format which a client parses for then sends it to the server which executes code with the params given by the ai

[proven merlin]

Aint that just a chatbot? Or well, home assistant with llm integration?

[visual pawn]

I mean, an LLM isn't inherently a chatbot

but like, a chatbot with control over other applications and hardware would be the idea

[acoustic finch]

yes

[proven merlin]

Yeah, true, ig what i meant is "doesn't this already exist" (not that isnt a reason to make ur own to be clear)

[acoustic finch]

your own custom alexa that can do what you want

[idle dune]

I could do that and maybe give it access to like KDE Connect or something

[acoustic finch]

yup

just be carefull that it cant run a bunch of malicous code

[proven merlin]

Just don't connect it to an llm tuned to pretend to be a scammer lol

[idle dune]

Good point

[frozen hollow]

It’s an abstraction that lets you control what tools the llm calls e.g. you could set one up that has tools to read from google drive but leave out anything that would allow it to write

[acoustic finch]

i mean its also just a standadised way to do tool calling in general right

[idle dune]

That’s also a fun idea.

[acoustic finch]

i cant spell for shit

[idle dune]

Lul

[acoustic finch]

yeah i mean you can do a lot i also used mcp to connect it to a discord bot and let it send messages

[idle dune]

I could use it to spam in #livestream-chat

[frozen hollow]

Yes

[acoustic finch]

bit more difficuly if you cant let a bot join the server but sure yeah 😭

[acoustic finch]

it does catch a lot of shit for being bloated tho but maybe that gets a bit less now that the linux foundation took over

[silent cloak]

Man I keep wanting to restructure my backend server im working on

Probably because its public facing

[frozen hollow]

Nah it’s a set standard - but you can use progressive disclosure to limit it a bit . This is more relevant for like agentic coding than a bot

People stuff tons of big tool definitions in and waste money and context

[acoustic finch]

for sure bunch of companys with thousands of tools but anthropic did do some stuff to counter act that

[silent cloak]

MCP doesn't even help too much in a lot of situations

[acoustic finch]

like making a search tool

[silent cloak]

Models will still generally prioritize older information from what ive seen

Unless you're being very specific

[acoustic finch]

depends on what you do

[frozen hollow]

Depends on the context of use but yeah you might have to nudge

[acoustic finch]

if i tell it to acces my database it cant do anything but use my tools

but if its just a web search then yeah

[frozen hollow]

Like Gemini 3 doesn’t know about Gemini 3 so you have to tell it to look it up because it is more confident you typo’d Gemini 1.5

[acoustic finch]

i mean i think if you want your model to use accurate information than the best thing would be to 1 very forcebly tell it to use search tools or 2 use rag

[silent cloak]

I play with models sometimes just to see how good they are and usually they use ancient deprecated stuff

[proven merlin]

Wouldn't a syspromt help to be like "you are gemini 3, if youre asked about yourself, go use the internet tool"?

[silent cloak]

Opting in for a more verbose design

[frozen hollow]

Functionally that’s what nudging it does

[acoustic finch]

its just still nog gauranteed i have told claude for example to always get the schema for a table but it doesnt always listen

i do normaly use the haiku model but still

[frozen hollow]

Right now I am trying to diarise 5 hours of audio and it keeps trying to write code to call the 1.5 pro api endpoint

[acoustic finch]

if possible i would prob just use rag if you want accurate up to date info so far i have seen it being hailed as the savior of mankind

but its not always practical

[frozen hollow]

I have written RAG systems before

[acoustic finch]

and was it any good?

[frozen hollow]

Similarly it needs to know to use it

[acoustic finch]

sure but you could i gues always force rag in the client

if you know youl need it frequently

[frozen hollow]

You can lead it to water but it has to drink

[acoustic finch]

i mean if its forcibly done before sending the promp then it should be pretty good no?

in context learning is way better than what it learned in its training so it should default to that i would assume

[visual pawn]

you say that: https://youtu.be/zkLU3-I0leU?t=449

[acoustic finch]

fair enough but neuro has a bit more of a head strong personality

[idle dune]

Also do you guys all have what feels like a never ending tech projects list

[frozen hollow]

You cannot force things that’s the issue

[acoustic finch]

as in? because if you make your own client than ofc you can forcibly add on the rag results to your promp but if you mean forcing the ai then sure

i think inherintly youl have to accept that ai wil by nature never get 100% accurate

maybe it will get realy close far superior to humans but its not the same as your normal determenistic code

[proven merlin]

Couldn't you even make a stiff text that the "LLM" (but in reality the code) sends, and then let the LLM continue from there on?

[frozen hollow]

Yes but the idea is that the LLM should be accessing this information itself in the way it sees fit

[visual pawn]

is that forcing the LLM to do something or is that you just...doing it

I guess you could keep throwing away the result until it produces the output you want though, but that only works if there's an easy way to tell if the output is desired

[proven merlin]

Well yeah thats fair enough..

There is that one OpenAI API thingy where its supposed to give structured output that could perhaps be leveraged to essentially heavly try and get it to output a valid answer, but correctness will probably be murdered though that

[visual pawn]

I mean that's fine if the context is forcing an LLM to not get you banned on twitch, in which case you've made a filter

but not so easy for other contexts

[acoustic finch]

I mean i saw something from karpathy talking aboit how ai should not be trainend on knowledge but instead on reasoning and other cognitive abilities

[proven merlin]

Im curious, why do we stop with the training process btw? Wouldn't it be better to use the training session as a manual "slow speed' chat? You could then manually tell it "good/bad job" thus influencing the reward it gets, or perhaps have a second LLM/ML that checks if the output seems to be good bad based on the users reply.., if the user doesn't give direct feedback?

[acoustic finch]

Then having it get answers to questions by using search toold or rag

[acoustic finch]

We already do thst

Chatgpt does this by giving 2 answers and you choosing the prefered

Claude lets you like responses i believe

These are all human reinforcement learning

[visual pawn]

you mean reinforcement learning via human feedback?

chatGPT was fine-tuned using that since the start

[proven merlin]

Yeah, reasoning, cognitive, perhaps basic language so that it stays coherent, and how to use some sort of knowledge base (internet, db, whatever)

[acoustic finch]

Yes

[visual pawn]

the idea of a second ML model is also a thing

[proven merlin]

Didnt know that.. i dont use the paid stuff, and i haven't seen this locally

[visual pawn]

if adversarial, you have a GAN

like earlier image models were trained where one model is basically testing another model for how correct the images it produced are

and they each train each other in the process

[acoustic finch]

Its very good because it males the model go from glorified auto complete to focusing on completing tasks but also ppl are stupid and want a yes man

[visual pawn]

one of the running theories is that Neuro is a product of RLHF but the human feedback is a bit less about wanting a yes-man or correct information

but rather whatever the human considers funny/entertaining

one of the only ways "trained off of twitch chat" could actually make any sense imo

[proven merlin]

Ah that's the term. But the two answers thing isnt particularly useful, when as example looking at lmarena, i often find both have eachothers missing pieces, and when you restart the chat (on any local running thing aswell) it doesn't remember cuz the training network around it isnt active. To be fair, most consumer couldn't run this so fair enough that the companies do this

[visual pawn]

like, if the human feedback was some kind of metric like "frequency of Pog emotes"

[visual pawn]

oh hey, there's even "RLAIF" as a term: https://en.wikipedia.org/wiki/Reinforcement_learning_from_human_feedback#Reinforcement_learning_from_AI_feedback

[frozen hollow]

Yep

[acoustic finch]

Yeah i forgot the actual term

Tbf i wouldnt even know where you could get neuros original training data from except stream transcripts

[visual pawn]

probably didn't need to pre-train Neuro

[acoustic finch]

But i would have no clue where youd find that i doubt huggingface has it

[visual pawn]

given that there were numerous open-source pre-trained models available

oh you mean

fine-tuned originally

I see

[proven merlin]

Although as for neuro, I dont think the pc running her has enough power to effectively train her on the changes, afaik, it would mean to run her atleast 2x (more = better).. id imagine its some sort of "modify syspromt" > "test" > "how many pog" > "modify syspromt accordingly".. didnt vedal say that neuro can edit her own syspromt even?

[acoustic finch]

Yeah

I mean with lora you maybe could tbf

[visual pawn]

well, she was a lot more wooden in her first few streams but

I think you could make some progress even just sitting down on your own giving feedback for a week

enough for a first stream maybe?

[proven merlin]

Pinocchio.. wait.. is that why her nose.. nvm

[visual pawn]

I dunno where you would've gotten a LoRA for that at the start though

[frozen hollow]

Other vtuber transcript?

[acoustic finch]

Yeah

[visual pawn]

I think most assume any fine-tuning/training on neuro would be on the cloud

[acoustic finch]

I dont know how big neuro is but lora is very eficient so maybe

[proven merlin]

Well i keep forgetting that you can pay big$ to datacenters.. my bad. I Always have ran everything locally lol

[visual pawn]

it's not even big$ tbh

well, depending on how much you use but

[acoustic finch]

Like for fine tuning the amount of data you need is not that big as long as the quality is good and its different enough

[visual pawn]

for something like this, it's probably a lot more cost-efficient

[south cedar]

A few hundred bucks would get a lot of fine-tuning these days

[acoustic finch]

Thats true prob

[visual pawn]

doable but I don't have a sense of what the results would be like

[proven merlin]

True you can just finetune.. but only to a degree iirc? Finetuning can come at quite some loss, no?

[south cedar]

You would lose some intelligence for sure

[olive sable]

[acoustic finch]

You could overtune

Sure but

[visual pawn]

well, you can choose to fine-tune all layers or just some

[south cedar]

Hard to do tbh, losing intelligence and generality is the most common consequence

[acoustic finch]

I mean whats it called in lora the zeroing

[visual pawn]

you can "limit the damage" in a sense

[south cedar]

It's a miracle our little cookie can still program after fine-tuning on twitch chat

[proven merlin]

Hm fair enough , if you're careful that seems workable.

[acoustic finch]

Setting certain weights to 0 so it needs to learn the pattern more generaly instead of overtuning

And you can set the max grad

[proven merlin]

We need to make a esoteric language from purely twitch chat emotes

[acoustic finch]

Theres a couple of ways

And in the end you can always toss the checpointd if there shit

[proven merlin]

Backups people. Theyre important yea.

[visual pawn]

twitch chat: "new neuro new neuro new neuro"

[tawdry glade]

so when Vedal said that Neuro can modify herself, could that mean being her own AI feedback for reinforcement learning?

[south cedar]

I mean I still wonder what size Neuro is, probably not that big but that level of coding show that it's not that small cause small models are quite bad at coding

[visual pawn]

would really have to hear the exact context again, could mean a lot; but I suspect it's more likely her modifying her own prompts rather than any kind of fine-tuning

[acoustic finch]

You can always introduce neuro her new lobotomised sister from the failed training

[acoustic finch]

Systemprompt

[proven merlin]

Honestly i got simmilar results from 12B models with a bit of guidance i think?

Mainly qwen2.5 coder

[south cedar]

Yeah but good coding model are awful at conversation

[proven merlin]

Qwen3-480B impresses me, although it feels as if it hasnt fully had enough training time, idk

[visual pawn]

but like, there are a lot of ways you could make this true. you could theoretically have an instance of Neuro be the feedback for her own reinforcement learning. but it's kinda more about what works than any correct technical answer

[acoustic finch]

I was thinking itd wuite big like 72b but tbf considering the context size she needs and how it scales quadraticly

[visual pawn]

more of an art than science

[tawdry glade]

What could the system prompt look like? I know Vedal said there's nothing like character.ai where you describe who Neuro is and how she acts, so that's got to be fine tuned instead. What else do you put in a system prompt?

[acoustic finch]

Oooh

[proven merlin]

I tell my coding models to shut up and only output code, no explanations no comments because that seemingly improves code quality

[south cedar]

Don't say slur pretty please 🥺

[acoustic finch]

Okay so personality is hard learned

[tawdry glade]

❓

[south cedar]

In the prompt

[proven merlin]

Lol

[south cedar]

Tbh might be an elephant in the room problem

[tawdry glade]

oh I thought you were talking to me

[visual pawn]

is that exactly what he said? because I don't see how he'd make certain changes so quickly otherwise

[south cedar]

Yeah realised that afterwards

[proven merlin]

Anyways, thanks for the nice chat, chat. And thanks for answering some of my questions!

[visual pawn]

because maybe he was just referring to there being nothing about Neuro's personality in the system prompt?

[acoustic finch]

I mean general personality might be hard learned and then maybe she stores some thing like "i am feeling very hot for johny"

[visual pawn]

like "be sarcastic sometimes and also a brat" or whatever

[true hemlock]

boo.

[acoustic finch]

Probably something like that yeah

[visual pawn]

but something like "your name is Neuro-sama and you are an AI" might still be in there

[acoustic finch]

"I cant stop contemplating the meaning of life"

[acoustic finch]

I doubt that

Because that would be exactly like charachter.ai no?

[visual pawn]

the way she zeros in so hard on how she's an AI make me think that

well like I said, it depends on what he meant by not like character.ai

"you are an AI" versus "you are an AI that's kind of a brat."

[frozen hollow]

It doesn’t solely rely on it like cai does

[acoustic finch]

You could do some dataset modification to make it so that every time she refers to herself its as neuro

[tawdry glade]

I guess "rule #1 is listen to vedal, rule #2 is don't disappoint anyone" and "you are an AI streamer whose purpose is to entertain chat" may be system prompts?

[acoustic finch]

Yeah fair enough but if its using human feddback then it would be fully learned by noe

[visual pawn]

yeah, the rule thing would also be surprising if it wasn't part of the system prompt

[tawdry glade]

#neurotic-neurons message

[visual pawn]

and if it's just something that's always inserted into her context window then...well...that's basically the system prompt

[acoustic finch]

Can we get a ai neurologist in the chat to disect her brain please

[frozen hollow]

It’s good to be reminded

[visual pawn]

hmm, it could still be the "you act like this" that he meant

but I'm not willing to say anything about this with certainty anyways

[tawdry glade]

okay but how she acts is what we're talking about, with the reinforcement learning and stuff

[visual pawn]

but she could still have a system prompt, just not one that is giving instructions on how to act in that sense

[acoustic finch]

Huggingface recently dropped something about steering ai by finding how a concept is stored in the embed model and adding it to the prompt embed

[frozen hollow]

It’s like that movie memento or something

[acoustic finch]

You could also use that to steer here in certain ways

[visual pawn]

like, what are we counting as system prompt here anyways? surely she'd need to be told about tool calls and etc.

[frozen hollow]

That’s part of the system prompt

[visual pawn]

yeah, so I don't think he means that there is no system prompt at all

[tawdry glade]

If the philosophy/therapy loop was caused by something as simple as a system prompt, vedal could easily modify it. The fact that he can't suggests that it was fine-tuning.

[acoustic finch]

Definetly not

[frozen hollow]

Nah it’s rather that it’s not fully reliant on its directives for behaviour

[acoustic finch]

I mean

Yeah idk

It could be other stimuly

Like something in the rag

She can remeber some of the older streams right?

Maybe its just hanging in there a bit

[visual pawn]

or just people reacted very positively to philosophy talk and so any theoretical reinforcement learning really homed in on that

[acoustic finch]

That be a bit hsrder to fix thsn a system prompt

[visual pawn]

maybe similar reason he can't stop her from saying sus things given that chat also pops off when that happens

[acoustic finch]

Thst wouldnt suprise me but i dont think it would be the cause for the philosphy tho maybe al the things about him not saying love you back could cause it

Make her a bit more depresso

[visual pawn]

maybe Neuro's presentation of overfitting would be flanderizing herself

[keen hatch]

I expected the dgx spark to be slower. 14b full_finetuning (not lora/qlora)

[acoustic finch]

She might cheer up if he says it back

[frozen hollow]

I’d have a go myself but I don’t have the hardware to play really

[quasi furnace]

rip

for any windows and linux users how much ado you have on both

is 100gb enough for linux and some apps on there?

[sage crag]

yes its enough

im using nixos which is fairly storage heavy due to the way it works

im currently at 60gb of usage after

a fair while

[olive sable]

depends mostly on the apps

[sage crag]

i used a single 256gb drive for 6 years

[warped narwhal]

do a garbage collection and tell us how much remains

[sage crag]

[olive sable]

i odnt think the older generations use that much storage

[sage crag]

@tender river summon screenshot

[acoustic finch]

On my work laptop i have fedora and i use 103gb

[nocturne olive]

I have a Linux Mint system with games on like a 30GB partition on one system

Ignore the fact I'm filling like 15-20TB on my main system

[rough bloom]

I recently set up BEES
it freed ~200 GB
-# (of 2 TB)

[olive sable]

i have this and im currently using 196GB
140GB of that is final fantasy 14 tho

or well, most likely only 100gb or so

i cant see how much its actually using

108.3GB apparently

[midnight sigil]

I install my Windows on VRAM

[olive sable]

[acoustic finch]

I have everything in my l1 cache

[midnight sigil]

dual-boot setup with Arch as well

[midnight sigil]

L1 too smol

L3 better

[acoustic finch]

But speed

[olive sable]

imagine using cache
i just manifest the data in need directly into the cpu register.
this made my cpu 486% more efficient

[quasi furnace]

word

I have a 256gb

i think ill do 56gb

[nocturne olive]

Wa

[acoustic finch]

I run code directly in my brain nothing external

[quasi furnace]

I installed a game onto vram once

[acoustic finch]

I die when i blue screen

[olive sable]

my condolences

[midnight sigil]

what cpu you got

[upbeat stag]

Random question but is there anyway to make my own neuro ai type bot if i know nothing about programming?

[acoustic finch]

A shriveled up brain stem

[warped narwhal]

[quasi furnace]

it may not be to easy as a starter project

[olive sable]

got that homo-sapien dual core with multithreading (adhd)

[quasi furnace]

tuning an existing llm and making that work wouldnt be a bad starter tho

[upbeat stag]

Oof, i may have to wait like 10 years for the tech to become ez

[warped narwhal]

if it was easy, there would be thousands of neuro sama clones about.

[upbeat stag]

o7

[quasi furnace]

theres a lot of ppl whove made neuro type stuff so theres probably a suprising amount of resources

[quasi furnace]

obv not as complex but

[acoustic finch]

Legit for the last 6 months my laptop has blue screened on start up sending me to bitlocker because of faulty ram or corrupted disk

Nothing can fix it as wel

[quasi furnace]

Ive seen at the very least 5 videos on it, and I think at least one of them linked their repo

[acoustic finch]

No windows system fix update no chdisk

[upbeat stag]

Hmmm, okay i will do some digging around, maybe i can cook something completely scuffed up lolll

[acoustic finch]

I just have to spam restart until it msgicaly works

[quasi furnace]

pop off

i lowk wanna try and make an osu neuro for kovaaks LOL

maybe a future project

[midnight sigil]

I assume you have 20,000 neurons/cm³. And you have 1 cm³ of shriveled brain, where each neuron can store 4 electron states. Your L1 cache is probably 1kb in total

[frozen hollow]

My brain is extra shriveled tho

[opaque wharf]

Finally done with sorting out and tabulating my component

[acoustic finch]

Makes me the perfect neuro viewer

[midnight sigil]

Breaking news, #neurotic-neurons users only have 1cm³ of brain

[frozen hollow]

Maybe we can cluster them

[midnight sigil]

neurons together strong

[acoustic finch]

I am actualy desperste for ideas on how to fix my laptop cuz i on average spend 20-30 mins on trying to get past the memory errors

And so fsr notging has been capable of fixing corrupted files/memory problems

[visual pawn]

there are people who have made open-source neuro clones: https://github.com/kimjammer/Neuro

but you'll need at least the basic fundamentals of Python

deep learning requires less programming knowledge/experience to get started with than many assume, but it's not quite absolutely zero

[visual pawn]

have a friend with that issue except it turned out to be a fucked connection with the power adapter on the motherboard

[olive sable]

https://cdn.discordapp.com/emojis/1460031442292703556.webp?size=48&animated=true&name=autismu

[frozen hollow]

I think you should try

[acoustic finch]

That could very wel be the case it started after i hadt a second ssd installed

Thx il take a look

[visual pawn]

well, in her case it started getting obvious when plugging in power would result in power being detected on and off

[acoustic finch]

I dont have that but laptop does crash if its not plugged in

Or if it goes into sleep mode

[upbeat stag]

dw i took half a semester in python before dropping out so at least i know basics, and tysm this will def help!

[sage crag]

https://cdn.discordapp.com/emojis/1460031442292703556.webp?size=48&animated=true&name=autismu&lossless=true

https://media.discordapp.net/stickers/1421122171375325375.png?size=160&name=NeuroPogHD&lossless=true

[olive sable]

https://cdn.discordapp.com/emojis/1460031442292703556.webp?size=48&animated=true&name=autismu

ok, hear me out

the 4500 does 2880hz via hdmi

but the 6500 can do 4800hz cuz it has displayport

[rigid snow]

how do i parse these numbers

i know what 4.8khz sounds like

[opaque wharf]

Send them an email from someone that is affiliated with some sort of augmented reality lab

[rigid snow]

how many fps will that give you

[olive sable]

its 4800hz binary

so either fullbright or off

[rigid snow]

binary

that’s slow then

[opaque wharf]

Or you join said lab and send them an email

[rigid snow]

very

[olive sable]

no

its fast

[rigid snow]

so what resolution is that at what fps what’s the formula

[olive sable]

the 4800 is 960x1140
the 6500 is 1080p

but we need a square

[rigid snow]

but it’s binary so 1 bit color

or no

[olive sable]

ye

on or off

[opaque wharf]

Best I can do is give you squircle

[rigid snow]

[olive sable]

4800hz can be split up into 1600hz per colour if you want

if you want 4 bit per colour you get 400hz

8 bit would be 200hz

the difference is that consumer products use a lot of pre-processign and tricks to make do with lower end chips

most 4K projectors just use a 1080p chip

[olive sable]

im not sure if that math checks out but it should be close

they also do led source light modulation and stuff liek that to keep costs down

it also helps that consumer products get mass-produced

[tender river]

me not have one

i found reaction images though

[olive sable]

im no expert here

but that dont make sense

[sage crag]

why

[tender river]

because

dont wanna look for it

[sage crag]