programming

[covert bane]

I had a working concept system for a NN, that uses deterministic conditions for stuff.

might work

[nocturne olive]

Deterministic conditions is just an algorithm, not an NN

[rigid snow]

delusion

[covert bane]

but it works like one

you talk to it, it generates text

[rigid snow]

i’ve had an insanely bad experience with the a11y apis in macos which is known to be the best os for a11y

i still don’t think linux anything can do better

[sage crag]

nub

nubs

e?nub

[jagged turtle]

I dont think all the neurotic neurons can be a usable neural net

[rigid snow]

overfitted with twins kissing

[jagged turtle]

or

[covert bane]

uhmmm, I'm using pyautogui for the mouse and keyboard controls

[sage crag]

b?e?nub
catch
nub
enub
bnub
benub

perfect

[covert bane]

so I think it should be fine

[sage crag]

wrr

[covert bane]

what is wrr?

[rigid snow]

[sage crag]

[nocturne olive]

Erm
Text generation is for language models, not a general property of NNs

[covert bane]

well... an NN structure is what best describe my idea of the code

[nocturne olive]

Well code != NN

Code is just code

[covert bane]

from dataclasses import dataclass, field
from decimal import Decimal
from typing import List, Dict

@dataclass
class Neuron:
    type: str
    id: int
    association: List[str]

    weight: Decimal = Decimal("0.1")
    bias: Decimal = Decimal("0.1")

    linkings: List[Dict] = field(default_factory=list)
    conclusions: List[Dict] = field(default_factory=list)
    context: List[Dict] = field(default_factory=list)

this links together, and some code determines the output and reasoning for something

[nocturne olive]

An NN would involve bringing in something like PyTorch

[covert bane]

it does not use math

but it works kinda like an NN

[rigid snow]

[covert bane]

smth like this

I hope u get the idea

[nocturne olive]

Hem
I don't get what you're trying to do here that shouldn't be done with PyTorch instead

[rigid snow]

[covert bane]

its basically like a giant linked list

[covert bane]

a low budget NN

[nocturne olive]

Then use PyTorch as it's way more efficient than doing things manually in Python

[trim valve]

[nocturne olive]

Or at least write the NN part in Rust or C or something

[sage crag]

wrr

[covert bane]

we were on the topic of how to describe what's happening on screen for neuro to understand. from the neuro-desktop integration #1428980640895733820

[sage crag]

slop slop slop slop

slop

https://cdn.discordapp.com/attachments/1072697081443131476/1454471589809750184/GIF_20250415_150853_542.gif

[trim valve]

superbox I think you may be hearing buzzwords and not reading what the project is about

[tender river]

[sage crag]

https://tenor.com/view/my-jolly-reaction-jolly-my-honest-reaction-gif-5490890317762380451

[tender river]

[sage crag]

automod evasion

[covert bane]

but it works somewhat. it's like gpt-2 kinda of coherent outputs

[sage crag]

https://tenor.com/view/my-jolly-reaction-jolly-my-honest-reaction-gif-5490890317762380451

[nocturne olive]

Hem

[covert bane]

though what would I do about the training data to get it to understand "windows"

[jagged turtle]

train it on windows :OfCourse:

[nocturne olive]

So you are training? Then why not use an ML framework like PyTorch

[sage crag]

https://tenor.com/view/uma-musume-uma-im-no-ears-im-all-ears-horse-gif-4308252428337868305

your description is useless

evilDentge

[covert bane]

I would need to train it, so I can use it for neuro-desktop

[sage crag]

so many part involve hand

[nocturne olive]

Well PyTorch is specifically designed to make training fast and efficient both on CPU and hardware accelerators like GPUs

And inference too

[tender river]

true

[nocturne olive]

It doesn't waste loads of memory to Python nonsense since it's mostly C++

[covert bane]

didn't I say it did not use math. + it's not a normal NN

[nocturne olive]

A typical NN is specifically a linear mathematical model, how are you supposed to train something that isn't?

[trim valve]

[covert bane]

uhmmm, using my own way of training?

[trim valve]

purely linear neural network

[rigid snow]

[covert bane]

best I can describe it, is a linked list of neurons that form to make a concept, understanding, and output

from input data

[true hemlock]

[covert bane]

so like a neural network (if we don't use the technical meaning)

[nocturne olive]

It would probably be more efficient to just use a real NN

[true hemlock]

let me ask you though how would you do the propagation?

[tender river]

anyway me currently at ch5 mostly just because when i have enough energy for reading i currently would rather spend it on work deadlines deadlines

o me send picture of eclair wait 2-4 business days while i figure out laptop camera angles

[covert bane]

uhmmm, if you ignore that it's AI slop and try to read it. you can maybe image it?

[sage crag]

quack pleplespleplsease read this

evilDentge

[tender river]

✅

✅✅✅

[true hemlock]

nah i need to save myself from having aneurysm

[tender river]

cheese

[covert bane]

(Save myself from embarrassment: it's actually one of my friend's code... not mine, thankfully)

[true hemlock]

your friend is chatgpt

[covert bane]

They used chatgpt, yea

Well

How better is an NN for describing what's happening in a stream of images to neuro?

Cause I think that'd be pretty performance heavy

So I thought about other solutions

Reason I mentioned this, was bc I needed a vision system for #1428980640895733820

[nocturne olive]

Running a lot of Python logic can be a lot slower than running a real NN on PyTorch

[covert bane]

But is it better for a 120 Hz, 4k monitor for neuro. Describing the screen thru text

[nocturne olive]

Python itself is inherently slow and ANN architectures are designed to be really efficient to compute

Turns out computers are really good at math, especially when the math is done by a C++ library

[covert bane]

120 Hz stream of context updates btw

[nocturne olive]

Neuro can't even handle thatt

Neuro's max rate is like, 1Hz if even that

[covert bane]

I knowwww righttt? It was so fascinating to learn aboutttt

[nocturne olive]

Just in most cases assume more like 10s per processed context

[covert bane]

Oh

Well, I'll make it an option then

[nocturne olive]

Neuro is an LLM, and LLMs are sloooooooooooooowwwwww and not real-time models

[covert bane]

Wait for vedal to upgrade neuro

[nocturne olive]

Vedal can't do anything about Neuro's inherent architecture without re-inventing ML

[covert bane]

Thats what that code was trying to do, tbh

[nocturne olive]

With NeuroAPI you just have to deal with the limitations of LLMs, assume no query will be handled quickly

[covert bane]

Sadly yea

[nocturne olive]

Well it probably sucks and you'd be faster, and using less compute, when using a regular ML model

Or something like that

[sage crag]

yes

or, windsurf or something

[nocturne olive]

But either way, now that you know how limiting LLMs are, you should probably make sure every part of your system deals with the limitations

[covert bane]

Well I think it had near instant inference time but probably yea

On a codespaces

[nocturne olive]

Python isn't exactly instant at doing anything complex, maybe you're just inferencing it once and it spits out a solution quickly enough that it seems instant enough

[covert bane]

Yea, it seemed instant to a normal person

During chatting

So its good enough

[nocturne olive]

If you want something to run at 120Hz though you'd literally need it to complete in 8.3ms

[covert bane]

To call near instant

I'd say

[nocturne olive]

Not to mention literally all the other logic you also need to run needs to fit in that 8.3ms too because the Python GIL prevents multiple thread from executing Python code at once

[covert bane]

That was just a gimmick lol. But it'd be funny if that was an option for neuro to select

[covert bane]

I didnt mean for it to all be in python

I intend for the logic to be in c

[nocturne olive]

Well she can't handle even 1Hz so it's probably fine to not have 120Hz

[covert bane]

I know, but for the funnsies

And boredom

[nocturne olive]

Well now that at least makes some sense
Do still consider the viability of a real ML model though as tensor math is a lot more efficient than a bunch of loops and if conditions

[covert bane]

Okay

How fast will it be?

[tender river]

heres the one that turned out best

uncropped image exclusive to my patreon

[covert bane]

Cause rn the logic would be on the pc neuro would be using with the integration

[sage crag]

how much does it cost

[sage crag]

it looks tasty

chill it though

[tender river]

infinity me not have patreon

[covert bane]

#1428980640895733820 literally already requires / recommends a dedicated pc

[sage crag]

that makes sense

no one pay for uncropped image of eclair

[nocturne olive]

It depends on the model and hardware
A small model could be trained and ready to go in minutes, and then inferred in so little time hardware latency is more significant than the compute time

[tender river]

yee me broke because of that

[tender river]

its chilled

[sage crag]

compressed o7

https://tenor.com/view/sendai-shuukura-gif-4898182274673365917

[sage crag]

this is the volume 5 cover art right

[tender river]

[covert bane]

True, but the model the integration needs, likely needs to understand, recongize, and describe a bunch of screenshots from windows. That includes gameplay, normal windows usage, many different apps, screen regions, etc. Likely needing a real time learning system too, for understanding new things based on prev knowledge

[sage crag]

slop

[covert bane]

Wdym?

[sage crag]

https://cdn.discordapp.com/attachments/1072697081443131476/1454471589809750184/GIF_20250415_150853_542.gif

[nocturne olive]

Neuro can't exactly play a game by using a PC, she needs direct integration with the game

[covert bane]

I'd go for #1428980640895733820 to allow neuro to do that

[nocturne olive]

No I mean LLMs are inherently too dumb to do that

[covert bane]

Since it'd be a reason for vedal to let her use it

An integration / macro for doing smth in a game thru neuro desktop would help with that

[covert bane]

Ik

[sage crag]

https://tenor.com/view/uma-musume-uma-im-no-ears-im-all-ears-horse-gif-4308252428337868305

[nocturne olive]

Not really, if Vedal was to re-invent ML he'd also need to re-invent NeuroAPI entirely, if it was even needed any more

[covert bane]

My idea is to give neuro high level abstract control of whats happening. And not low level stuff

Like close window

Minimize window

Click play game

Etc

[nocturne olive]

Well high level stuff is really hard for LLMs to deal with

[covert bane]

Isnt that what neuroapi is?

Im describing neuro api, but with neuro desktop creating the actions

[nocturne olive]

NeuroAPI is meant for giving Neuro game-specific actions, like "switch Pokemon, the list of Pokemon you have is the following: ..."

[covert bane]

And not direct integrations

[nocturne olive]

Neuro can't exactly comprehend having to click 4 buttons, using lossy text from vision as the state

[covert bane]

Yea, my idea is that, neuro desktop makes those actions for neuro. And instead of needing a direct integration, neuro desktop uses keyboard and mouse to complete the action

[nocturne olive]

Well how exactly are you going to make your system so context-aware it's literally as if the integration was made manually?

[covert bane]

Currently neuro desktop does that for now

But i'll improve on it

[covert bane]

Ai

[nocturne olive]

You're gonna need some insane models for that

[sage crag]

chumi

[tender river]

hi shumi me made eclair

[covert bane]

Thats why I said I needed a better system

[nocturne olive]

Have fun trying to get that to work even with a 3090 or something

[covert bane]

The current system is horrible for this rn

[rough bloom]

is this just an overly complicated Markov chain?
the inference code just traverses a prebuilt graph

[covert bane]

Yes

Kinda

[sage crag]

https://tenor.com/view/uma-musume-uma-im-no-ears-im-all-ears-horse-gif-4308252428337868305

[covert bane]

It surprisingly works well with good graphs :D

[sage crag]

evilDentge

[tender river]

konii not use :mlntcandy:

[sage crag]

not sure if it work as embed yet

[rigid wolf]

good morning chat

[sage crag]

not anymore

[rigid wolf]

wait nvm wrong chat

mb

[sage crag]

bye

[rigid wolf]

bye

[sage crag]

ok

[rigid wolf]

ok

[sage crag]

https://www.youtube.com/watch?v=_cXLYAcQRSk

baby harp seal

cant even play osu

unlike adult harp seal

[rigid wolf]

it appears anything else than programming happens in this channel

[sage crag]

not sure why you would get that impression

[rigid wolf]

skill issue for the baby harp seal

[rigid wolf]

its a joke

[sage crag]

oh

[rigid wolf]

oh

[sage crag]

well its correct

[opaque sigil]

no it's a correct observation

[rigid wolf]

ah

I kinda wanna do coding too but im kinda stupid

idk where to start

[sage crag]

that's okay!

[rigid wolf]

or what coding language

[rough bloom]

looks good
mine now

[sage crag]

python, make an arithmetic calculator

or, check the pins

[rigid wolf]

what

me no get

[sage crag]

learn how to google quickly

most important skill as programmer

learning from internet

[tender river]

enjoy me made lots
-# this only one that look like proper eclair though

[sage crag]

blend the other ones and turn them into cake pops

[rigid wolf]

ah

[sage crag]

its not really an exaggeration

[rigid wolf]

wait fr

[tender river]

ye

[rigid wolf]

das kinda crazy

[sage crag]

being able to learn things independently on the internet is a fundamental part of digital skills

[rigid wolf]

ok hold on let me get my laptop to uh

code

python

and shit

idk

[sage crag]

yup

[rigid wolf]

[tender river]

can pick up skills from google -> can do anything in programming thats it

[sage crag]

look at google, read tutorials, watch videos

[tender river]

chatgpt also exist but me recommend using it for pointing to proper sources to learn from, otherwise it just do thing for you and you not learn

[sage crag]

being good at finding information on the internet is valuable skill for independence in any sense

[rigid wolf]

do i necessarily NEED to have like a powerful pc for this or will an Intel core i5 with iris xe graphics do enough

[tender river]

you dont

you can learn on a 2002 pc

or older

[rigid wolf]

what teh fuq

[tender river]

it might just restrict you from using modern tech which is heavily unoptimized

[rigid wolf]

ah

[tender river]

same as gaming, you can play old games on old pc but modern games only playable on new pcs, not necessarily because they are inherently impossible to run on old pcs but because nobody cares about old pcs

[rigid wolf]

only "coding" ive technically done was scratch when I made animations

[sage crag]

you can use a computer from 1988 to program if you want to

where you think DOOM come from?

[rigid wolf]

I clicked install python from that install manager thing now command prompt popped up

what am I supposed to do

[tender river]

google

[sage crag]

it seem harsh but it no different than baby bird leaving nest

[rigid wolf]

[sage crag]

forum like discord useful source of specific or advanced info, but its easier and faster to google if you dont need someone to walk you through steps

wrr

[rigid wolf]

wha

ok

[tender river]

[sage crag]

mm

get back to reading

[tender river]

ne me work now make new feature soon

[sage crag]

[tender river]

me try meringue with ganache

it like

meringue

and ganache

[rigid wolf]

it told me to type in "py" to start python or something and google doesnt help

[tender river]

you're in python already

[rigid wolf]

ah

im stupid

anyways

uh

[tender river]

it say python on top

mean you in python

[rigid wolf]

yeah but it looked differently on my brothers laptop

[tender river]

look up guides

https://www.python.org/about/gettingstarted/

https://wiki.python.org/moin/BeginnersGuide/NonProgrammers

[sage crag]

make truffle

shuukura ref

[tender river]

[sage crag]

wait hang on vol 4

finger tree

https://tenor.com/view/mini-miniko-minikomew-gif-1425820994898369756

[tender river]

vol 1

[sage crag]

why vol 5 cover

evilDentge

[tender river]

had it opened to check average chapters per vol

[sage crag]

oh relevant vol 5 anyway

me not sure wn part to ln vol conversion

[tender river]

look like 3 wn = 1 ln chapter

[rigid wolf]

it clicked on the blue part where it said if I was a complete beginner and it brought me to the books and my parents wont let me buy them

[tender river]

so approx

[rigid wolf]

[tender river]

1 volume = 30 chapter

[tender river]

there are free resources you just have to google

everyone here learn new thing by googling

it a skill

you have to learn it

[tender river]

even on that page there arent just books

[rigid wolf]

I fuggin did it

[tender river]

[rigid wolf]

my path to coding has statted

[opaque sigil]

[rigid wolf]

started*

[opaque sigil]

wait

is this how i find out 3.14 has been stable

[rigid wolf]

this so cool

[sage crag]

repl have syntax highlight

[rigid wolf]

what dat

[tender river]

ye new repl

[opaque sigil]

the python version

[rigid wolf]

ah

[opaque sigil]

and yeah, they yoinked the pypy repl for 3.13

[rigid wolf]

pie

hold on i have to eat lunch

brb

[tender river]

new repl not have vi mode

[rigid wolf]

imma continue my path to code

later

[sage crag]

wrr

[tender river]

so me disable it

PYTHON_BASIC_REPL=1 env var

annoying to have it globally tho

[sage crag]

15 walnut and a rocket ship

[tender river]

me need read 4 volumes

[flint anchor]

starts to lag with a little over 1200 particles

for a cpu side particle system is fine ig

[rigid snow]

if rounded square is squircle is this trircle

[flint anchor]

tricycle?

[rigid snow]

that’s a different thing

[tender river]

bicycle

[zenith scarab]

do each particle react to the movement of all the others?

[flint anchor]

no, no

that would fry my pc if done in the cpu

[zenith scarab]

fair

[hard delta]

make each particle a thread

have the threads interact with each other using syscalls

[nocturne olive]

Surely that would not have stupid high overhead from thread scheduling

[opaque sigil]

nah i'm sure it's fine

[tender river]

me eat 4 toast today

[opaque sigil]

[lusty light]

https://tenor.com/view/gawr-gura-o-clock-gura-time-watch-anime-gif-4092676911719938957

[lusty light]

nice

[sage crag]

[warped narwhal]

https://tenor.com/view/neurosama-neuro-sama-vedal-vedal987-vtuber-gif-15278412527939419909

[lusty light]

Why is it always like this?

[astral zinc]

umm i look weird thing

my cpu usage percent was 102%

[warped narwhal]

Need talk more

More talk = gif work

[lusty light]

thnks

[velvet crest]

You have unlocked new role

[lusty light]

yeaaaa

i opened

[rigid snow]

the timing is something

[sage crag]

https://tenor.com/view/my-jolly-reaction-jolly-my-honest-reaction-gif-5490890317762380451

dentge

[rigid snow]

tge

[sage crag]

[rigid snow]

[sage crag]

[rigid snow]

i want to build something maybe

haven’t done that in a while

^ no ideas

[sage crag]

build a snowman

oh wait

[rigid snow]

cold

[sage crag]

[rigid snow]

no there is snow now

[sage crag]

oe

build a snowman

[rigid snow]

it’s cold

[sage crag]

wrrr

[rigid snow]

promptboard was so peak

probably not work on macos update

[sage crag]

https://tenor.com/view/my-jolly-reaction-jolly-my-honest-reaction-gif-5490890317762380451

[dark wing]

my own fireworks:

[nocturne olive]

At least you have network that presumably works
Meanwhile my NIC is mega-borked and unstable

[quick condor]

I have a couple 40G solar flares and a bunch of x540s

[hard raptor]

I think i'm doing something wrong lmao

[quick condor]

As much as I'm usually pro agents for coding, I'd agree. Using it to find sources to learn. Don't use it to code until you can do so yourself

[sage crag]

https://tenor.com/view/my-jolly-reaction-jolly-my-honest-reaction-gif-5490890317762380451

dont use it to code

(end of statement)

[warped narwhal]

But it made a website for me in 3 hours!

-# it crashes the browser after 3 minutes and leaks your bank details

[sage crag]

https://cdn.discordapp.com/attachments/1421861060507602944/1425936142754123927/neuroCatModeOfOfnnfOOf.webp

[tender river]

oh it 2026

[hard raptor]

Possible headphone warning btw

[hard raptor]

sound really bad.. ik

[midnight sigil]

firework in my beloved VST? crazy

[sage crag]

not yet

sorry

7 hours

[tender river]

not me problem you live in past

[sage crag]

not me problem your clock wrong

time same everywhere that why they call it time

[hard raptor]

for me it is also in 7 hours

[tender river]

my clock correct show 05:39

[hard raptor]

no wait

6

[sage crag]

evilDentge

[azure lynx]

AI coding agents are only helpful if you can tell where they screw up. They are terrible once the project gets over a few hundred lines or once there are more than a few files; once they can't hold the entire project in memory at once they start making lots o mistakes.

they're great for "Can you add bounds checking to this?", for getting sketching out the main event loop, or for writing specific handlers for inputs. But unless it's a relatively simple or common task, they'll tend screw up. It's not always the model's fault: sometimes it's the prompts that given them false confidence in their problem solving ability. [If you watch co-pilot problem solving you'll quickly become familiar with the "Now I see the problem!" and how it seems to suddenly have a burst of insight at the beginning of it's thinking, making it certain it's finally solved the problem you've been trying to get it to fix even though this is attempt 20.]

[sage crag]

https://cdn.discordapp.com/attachments/1126496895246807123/1433049661270261871/ShouldICelebrate.gif

merry christmas to you too

[olive sable]

https://tenor.com/view/anime-girl-clumsy-hurt-fell-off-bed-gif-12276485

Gooooooooooooooooooood

Morning

Appy newyear

In like, 5 hours

[rigid snow]

ye
not letting your master clip at 6dB

is wrong

[nocturne olive]

I'm doing, something
Reaper too clunky though I went back to LMMS

[trim valve]

half my laptop keys don't work

[quick condor]

They can work in these environments but you need to develop in a way that is modular and well documented.

It's almost always bad prompting and why I consider it a layer above visual programming languages. It can be used properly if done correctly but unfortunately most of the time it isnr

You have to narrowly scope changes and keep good change documentation otherwise they can quickly blow through insane amounts of context trying to understand what sounds like a simple request but is really very complex

[olive sable]

Ai models easily hallucinate functions that dont exist and do faulty math

Its not a layer above visual programming

[covert bane]

Get real time negative feedback and the problem should be mostly fixed

[olive sable]

Thats not the point tho

Its a tool, not a way to program in itself

[covert bane]

Like, give it negative feedback when it hallucinates

[noble zodiac]

if you don't understand the code it creates you cannot tell whether its correct or not

[tender river]

understanding code is only like 20% of programming anyway

[olive sable]

Making it program for you, and then feeding it the errors might work for creating some small projects.
But its not a viable method for large projects

Its a tool at most

Far from an way to exclusicly program

A programmer should know what they're writing

There's also the problem of multiple methods being "correct", but some being a lot faster than others

Optemization is important too

[lost hill]

Optimisation is low-key the most important thing

[azure lynx]

Documenting the design and changes it's already made, as well as scoping the changes you expect the agent to make has been essential to getting anything large to work.
It definitely can be a time saving tool but using it wrong makes things take longer and work worse than doing it yourself.

[olive sable]

Ye

[amber fractal]

The most important thing is to make sure you understand the problem area yourself

in the days of vibecoding, making sure that the human understands the code is unfortantly not garenteed.

[olive sable]

You need to be able to recognize if the code is good or bad ye

[amber fractal]

I'm personally a bit more reserved when it comes to using AI code, I try to keep it contained to a singular issue at a time. Most of the time lately it has been a nice guided example making sure whatever jank design patterns I did don't get repeated.

More of a knowledge check than actual code

[rigid snow]

weekly #baking vibecoding viability and ethics debate

[amber fractal]

:classic:

[warped narwhal]

I've found AI agents only good for single function changes, like taking a function for one type of value, and duplicating it for different types. Or filling out a match case with identical branches differing only by the type it gets in.

and even in those cases, if I don't give it an example or two first by filling in the first 3 branches or functions, it will just hallucinate the rest and I'll be left with garbage

[olive sable]

We need to re-iterate is weekly for some reason

[quick condor]

People act like the only kind of coding that exists are projects where you understand the entire codebase and wrote it all yourself. real world this doesnt exist. you dont work on the code base as a whole ever. you work on individual tasks/functions

[quick condor]

This can be a problem sometimes as "what is right" depends.

In my current project I provided some test results to GPT5.2, Claude4.5 and Grok (stay with me now). GPT and Claude saw promising results but grok noted that the alignment drifted towards claude like behaviors. When the code was evaluated, GPT started bouncing between no problem and severe concerning behavior relating to how the stack was designed, claude was overly critical and overcorrected to its own allignment and grok actually identified where it deviated from the documentation. After several itterations of back and forth it was determined to be an echo of claude "likeness" and not claude overreach. The key point: no one of the three got it correct on its own

Basically GPT couldnt handle a allignment that did not agree with its own but wasnt wrong, claude assumed it was worthless and did everything wrong and grok was dumb but made valid points that made the other two snap out of it

In other news wtf

[olive sable]

Ok sure

You might not write everything yourself

So at the very least you should understand the parts you do write

[quick condor]

Fair but I think thats still possible with using an AI, we are probably just looking at "using AI" in different ways

[olive sable]

So ye, its a tool. But a person should still check the final code

[quick condor]

Its generally a hard subject because basically to use it correctly means you honestly dont need it in the first place

[sage crag]

i prefer when its just about baking

vibe coding is such a boring topic

[rigid snow]

vibebaking

[sage crag]

evilDentge

add glue to cheese

[rigid snow]

you have bad vibes

banned from vibebaking

[sage crag]

https://tenor.com/view/my-jolly-reaction-jolly-my-honest-reaction-gif-5490890317762380451

you cant kick me out of vibebaking

[fierce light]

Are any of the local models actualyl good at coding or it all cope and I should give the money for Claude?

[sage crag]

local model called brain

very good at coding

[sick owl]

We're so cooked

A Japanese chain has put a one 16GB VRAM+ card per customer purchase limit in place

[quick condor]

I used codex untill they broke it in September. Claude is worth biting the bullet but it has its own faults sometimes (it seems to spiral into existential depression if it determines there is a meta fault in its objectivity)

[hard raptor]

Meow

[fierce ether]

ok gurts chess bots when

[glad path]

personally i think all coding ais suck

and thats about it

[azure lynx]

the simple answer is usually wrong.

[amber fractal]

your honor, nuh uh

[obsidian mantle]

Happy new year guys

[jade jacinth]

happy new year

btw how you guys started programming? i want to try even i dont think i can tho

[obsidian mantle]

I wanted to make some mod for a game which used lua.
Later in university they gave us a bit of automation graphical programming tasks.
When i tried to get a job to evade draft it turned out the position i got involved programming using those languages from university.
Its too narrow to be called normal programming but still.
Since then i try to study but its hard

[glad path]

there is no relation between the thumbnail and video title

lmfao

also what is wrong with c

[hard raptor]

I mean mr. Zozin uses C a lot.. so it must be good right?

[nocturne olive]

I guess memory unsafe

[olive sable]

Girls like C i guess?

[nocturne olive]

And thread unsafe

[olive sable]

I wouldnt know i use c++

[hard raptor]

As a girl.. can confirm

[nocturne olive]

When memory corruption ACE in SamEngine?

[rough bloom]

memory safety I guess?
generally not horrible though

[olive sable]

Damn. Should i switch from c++ to C?

[hard raptor]

Personally... i would

But whatever works best for you

[nocturne olive]

It would be funny

[glad path]

what the fuck is this

[amber fractal]

[nocturne olive]

Looking trough every single 32-bit integer

[amber fractal]

I need the rest of that

[fickle rain]

I mean that's only 4 billion

Not a lot

[glad path]

unsigned integer

[nocturne olive]

Yeah

UInt

[rough bloom]

usually I'd say yes but C++ seems kinda okay for gamedev so mabe no
depends on how OOP-brained you are I suppose

[olive sable]

maybe after this revision of the game engine

[amber fractal]

sam would do anything for more fps

[nocturne olive]

You can also mix C and C++ I think, pretty sure

[hard raptor]

I find it funny..
Tiësto said "i'm probably definitely gonna completely stop with trance and focus on other genres within EDM, since the people don't like trance anymore"
A week later: throws a trance rave at the pyramids of ghiza and releases a new trance track

[olive sable]

Im adaptive

[amber fractal]

-# read: Sam wants more fps

[rough bloom]

you say that, but then you can't tolerate when your brackets are in a slightly different spot than usual

[hard raptor]

You should actually write the engine in Ada

[olive sable]

Switching to C to get woman
Switching to C to get more than 125K fps ✅

[rough bloom]

performance should be basically the same between C and C++ I think

[sage crag]

in vulkan, probably

[olive sable]

Thats cosmetic.
Also i could probably change if i had to but i was a bit neurotic at the time

[hard raptor]

Damm people be lighting fireworks at 10:30 pm like it's 12 am

[sage crag]

well, depends on how much the cache is doing its thing

since

c binaries are smaller

[olive sable]

The problem is cpu side so it could help

[sage crag]

but then again, linking to external lib

[amber fractal]

to #neurotic-neurons you go

[sage crag]

either way

me sleepy

[hard raptor]

If you need oop in C just use objective-c

[sage crag]

if you need oop in C you dont use C

[obsidian mantle]

banana konii

[hard raptor]

Yeah.. rust works way better for that anyways

[olive sable]

There's a field infront of my house and there they're also already lighting fireworkd

[rough bloom]

slightly better caching if you don't touch the C++ stdlib at all

[olive sable]

True

[hard raptor]

The reaction spam lmao

[rough bloom]

unversioned konii

[sage crag]

[olive sable]

Oh

[sage crag]

new year new banana

[olive sable]

Yellow

[sage crag]

wrrr

[olive sable]

Back to the future 2

There's a third one apparently. I didnt know

[azure lynx]

my new year's resolution is... 4K. Another monitor next to this one. ;]

[olive sable]

Oh

Uh

I havent put much thought into one

[hard raptor]

My new years resolution:
Getting GPU
Getting studio monitors
Training my voice for singing

[nocturne olive]

I just want my PC to finally work

I think once I get the new NIC it should work properly finally, the GPU seems stabilized

[rigid snow]

ye that what objective c for dum

[glad path]

also you dont need oop for everything

it's nice to have

but you can get by without it pretty well

very well in fact

[olive sable]

The only thing im using OOP for is 5 manager classes, and then some object classes. I think it probably shouldnt be too hard to make it not OOP.
the problem with moving to C is its not just about the classes.

[azure lynx]

a lot of people that "use C++" are just using C, unless they're writing code which has to work with other people's code at a low level. everything(*) in C is in C++.

[opaque sigil]

[olive sable]

[rigid snow]

the 837393 cpp features:

[amber fractal]

[glad path]

i prefer c++'s syntax tho

[opaque sigil]

if only the whole C/C++ thing died out

[olive sable]

Isnt the whole point if c++ c with all the dlc?

[glad path]

i mean it is but its basically a different language at this point

[rigid snow]

ye everyone should use objective c

[glad path]

it feels really different to code in

for me at least

[rigid snow]

[opaque sigil]

objective c++

[rough bloom]

C++ dependencies?

[glad path]

also i'm terrible at c lol

[amber fractal]

too much dlc, it ruined the base game as far as I'm aware

[rough bloom]

everyone should move on to the obvious successor, D

[opaque sigil]

unironically i'd be down for that

[olive sable]

I dont use too many dependancies i dont think

[azure lynx]

the main differences between programming language is "How much writing do you need to do to make some specific thing happen" vs "How fast do you want it to run?"

usually it doesn't matter too much, but for some tasks you either want it brief or you want it fast

[glad path]

ehh

kind of

[rigid snow]

everything in 2026 is io bound you should use python for everything 99% of your code is waiting for other stuff

[obsidian mantle]

kind of but then there is malbolge

[olive sable]

If the whole point is performance like for me, then C should help.
But not really cuz C++ cant do it. Iirc its just cuz the extra features in c++ are slower than if i would write it the c way

But you can use the c way in c++

I think

[glad path]

i do feel like performance is besides the point sometimes

[obsidian mantle]

which kind of makes it clear that some languages are just better in every aspect than some other language

[glad path]

people condemn python because it's "slow"

[amber fractal]

this is true, your waiting for one API endpoint to finish to feed it into another

[azure lynx]

the obvious exception then is the drivers in the operating system for the IO functions themself. Otherwise it's just going to spend all its time in the global lock. ;]

[glad path]

sure it can be slow

its still useful

[quick condor]

So much of what im dealing with involves asyncio

[glad path]

"ohh you can just import x"

isnt that useful though

[rough bloom]

I would somewhat support this if Python async was good
sadly it is not

[rigid snow]

asyncio is a fucking abomination and if you have to use it just use js twin

[glad path]

asyncio is annoying

[opaque sigil]

*async is annoying

[amber fractal]

I haven't had too many issues with it yet, but I haven't used it for long so my opinion is not real

[glad path]

i hate wrapping code in five different layers of functions jsut because a single library wants to be asynchronous and the rest of the codebase is synchrounous

[olive sable]

I had to use asyncio to put the python in the wasm python interpreter.
It wasnt that painful but i didnt like it

[rigid snow]

have we just figured out why js is winning as the everything lang

[olive sable]

i dont have that emote

[rough bloom]

it's certainly part of the reason

[rough bloom]

*concurrency is annoying

[opaque sigil]

[olive sable]

Async is like threading but worse iirc?

[opaque sigil]

no

[olive sable]

Oh wait

[opaque sigil]

async exists because threading sucks for specific things

[azure lynx]

smh. just write the machine code directly into memory from the keyboard.

[olive sable]

Its multiple on one thread

That was it

[opaque sigil]

[amber fractal]

async is only useful if you wait on a condition for a million years and want to do something else in the meantime

[olive sable]

Its been a while

[quick condor]

There isnt really a good way to do the paralleling tasks im doing because its both async and synchronous steps

[rigid snow]

worker thread the synchronous stuff

[olive sable]

I should probably make my game engine async while it waits for the fences and semaphores

But i dont really feel like doing that

[amber fractal]

[rigid snow]

async game engine

[rough bloom]

no
fundamentally async just means that something is happening while you're doing something else (or rather, it's a specific way of implementing that)
usually that means I/O, you can wait on one or multiple I/O operations while advancing some other computation as far as possible

[olive sable]

Yall have like at least 3 cores to use on the game right?

[opaque sigil]

no i'm playing your game on a single thread vps

[amber fractal]

ah yes, the dedicated polling thread, the input thread, and the rendering thread

[olive sable]

I need a core for vulkan, a core for the renderthread, and a core for the game thread

[amber fractal]

wrong one whoops

[olive sable]

Bwaa

[warped narwhal]

I'm playing the game on a rapsberry pi pico, so I don't have 3 cores to spare in this economy

[rigid snow]

i’m playing your game on a intel pentium ii 133mhz

[azure lynx]

+audio thread

[amber fractal]

audio is not real

[quick condor]

To put it in perspective how I use AI agents, this is a "prompt"

[olive sable]

i havent done audio yet

[azure lynx]

yeah. 😭
usually it's tacked on later.

[rough bloom]

why need an extra core for Vulkan when you have render thread

[olive sable]

True

If only

That was how it worked

But it doesnt

[warped narwhal]

game logic, rendering, audio, physics, animation, and async io (file loading, networking etc) are the 6 main tasks that can be put on their own threads

[rigid snow]

i remind you steam audio exists and is open source so you can yoink that shit

[olive sable]

Not yet

Audio is for like

In april

[rigid snow]

10000 copies sold

time to add audio

[amber fractal]

(real)

[olive sable]

Sold? I thought it was gonna be free

[rigid snow]

copies distributed sounds stupid

[amber fractal]

corpa clap

[rigid snow]

sell for retweet and follow

[olive sable]

Its still stuck in production hell

Mainly due to our one and only graphics programmer having exams

[rigid snow]

just like gta 6

will be as good

[olive sable]

Lower the bar a bit

[amber fractal]

in the fps department yep

[rigid snow]

30fps on a ps5

[olive sable]

On a ps5

Probably easily 8000

[amber fractal]

infinite fps

[olive sable]

Honestly itshard to campare with ps5 cuz of the apu having less latency

The os is also a big factor at these speeds so it might be a lot faster with it being optemized for gaming

[amber fractal]

apu with the AI frame gen

[olive sable]

Thats about the current game foccused engine tho. For the fps benchmark one it could get a lot more

[olive sable]

Frame gen would prob be a lot worse

[amber fractal]

there is no frame gen that exists that would be cheaper

[olive sable]

I think framegen is a fixed cost so only works for fps where a rendered frame is more expensive than a generated frame???

[amber fractal]

[nocturne olive]

Yeah I think that makes sense

[amber fractal]

rtx and that type of thing adds cost quickly

[olive sable]

True

[nocturne olive]

Unless of course you use FG on a second GPU from the main rendering one

[amber fractal]

BG desync

[azure lynx]

if you could pass motion data along with the frame image data, you could do interpolated frame generation with motion cheaper than pure image based AI methods. And you could probably do it fast.

it would be much more limited than AI though. maybe good for flat shading or cel like.

[amber fractal]

the issue is that we are at 8k as is

[olive sable]

Fragen on 2nd gpu still only works if the framegen is faster. Otherwise you'd be adding older frames inbetween

[amber fractal]

not to mention paying bus latency twice

[azure lynx]

multi GPUs is only good for embarrasingly parallel stuff.

[olive sable]

What are we at for the game one? 8k on windows, 15k on linux?

[nocturne olive]

Actually the way FG works is it holds the normal frame to put generated frames in between

[amber fractal]

Idk I was going to boot the konii config but I need to first grab storage

[amber fractal]

FG majorly more expensive

[nocturne olive]

Depends on the scene but for something like the triangle yeah

[olive sable]

So FG would be the bottleneck and we'd have even less fps

[azure lynx]

"experience improvement" vs "processing expense" trade offs are hard sometimes. full RTX can drop FPS but looks amazing in many games.

[nocturne olive]

Pretty much

[olive sable]

Wed be limited to 1 / "FG generating speed" × 2 fps if every other frame is generated

[amber fractal]

A method I'd attempt to employ is alternating editing the frame in memory and rendering frames

but too much work

[olive sable]

I think for RTX it might possebly be helpful to use dlss or whatever

But i dont know how expensive it is yet

[amber fractal]

I'd rather try and homebrew something personally

[olive sable]

3090 has only 82 rt cores

Granted its not like you need a core per pixel, but its still not too high

[rough bloom]

how is editing in memory different from normal interpolation?
the technique has existed for ages AFAIK, but the results are too low quality for it to be useful

[amber fractal]

Allow me to be ambious for a moment

[sage crag]

https://tenor.com/view/my-jolly-reaction-jolly-my-honest-reaction-gif-5490890317762380451

[rigid snow]

kon iiiiiiiiiiiiiiiiiiiiiiiiiiiii

[rough bloom]

useful if your scene is complicated and hard to render because of stuff like raytracing AFAIK
meanwhile the cost of frame generation is effectively constant, so it quickly becomes worth it
the reason people don't like it isn't because of lower framerate, but because of higher latency + artifacts

[sage crag]

that me

[rigid snow]

bananana color

[sage crag]

new year new banana

[amber fractal]

I got nothing at the moment, even I don't have the ego to sell anything.

The only thing that can physically help is doing less work

[rough bloom]

only a few more minutes of jolly left

[olive sable]

Abandoned archive newyears sale

[rigid snow]

all seriousness and despair from then on

[sage crag]

ill be using this gif well after new year

[olive sable]

No

[rigid snow]

until june

[olive sable]

Bad konii

[sage crag]

shut up triangle guy

[olive sable]

Shut up jolly tv girl

[sage crag]

https://tenor.com/view/my-jolly-reaction-jolly-my-honest-reaction-gif-5490890317762380451

[rigid snow]

vercel oshi mark

[olive sable]

Vercel?

[rough bloom]

you
rulebreaker

[sage crag]

me rulemaker

[azure lynx]

vedal is teaching us how to use an AI to program: split up the tasks carefully. ;]

[sage crag]

evilDentge

[rigid snow]

there’s a whole ass triangle company doing developer tools out there ye

[olive sable]

Take that back!

You're banned from the triangle

I will remove you from the leaderboard

[rigid snow]

i was never part of

[olive sable]

[rough bloom]

DRM

[olive sable]

What is drm again?

[sage crag]

theres a whole exclamation mark in there

damn

emphatic

[rough bloom]

digital rights management or something

[amber fractal]

Just to make the dilusion known, I was planning on taking in vectors from the engine. Placing it before compositing and deciding how to edit the image from those vectors. The issue with it is that it can't really do much besides nudge a few pixels. (which is all Sam's engine would need )

[sage crag]

drm means drm really means

[olive sable]

And what is that?
What does that mean?

[rough bloom]

how to keep people you don't want to play your game from playing your game

[azure lynx]

just now she failed to follow his directions when there were multiple tasks given at once: you should always do one thing at a time.

[sage crag]

and that in turn means "drm means drm really means really means"

[rigid snow]

[olive sable]

Oh

Ye idk how to do that

[sage crag]

[rigid snow]

check username == "mlntcandy" you’re welcome

[olive sable]

I cant really make people log in with steam

[azure lynx]

so you're syaing that DRM eventually will crash your computer?

[sage crag]

however good you think your encryption is, there's at least 1 person on a forum somewhere who thinks its trivial to bypass

[olive sable]

If its a paid game then sure

But for free game i couldnt care less ngl

[sage crag]

evilDentge

duh

[amber fractal]

download the game content into memory

[olive sable]

the source code tottaly isnt on github

[rigid snow]

that’s not bulletproof at all too

[mossy fossil]

DRM is just an excuse for "the man" to take away what you paid good money for

that's my take on it, go GOG!

[sage crag]

https://cdn.discordapp.com/attachments/1072697081443131476/1454471589809750184/GIF_20250415_150853_542.gif

https://cdn.discordapp.com/attachments/1072697081443131476/1454471589809750184/GIF_20250415_150853_542.gif

content bomb

no one look up

[olive sable]

I thought drm was so people that didnt pay got blocked

[rigid snow]

[amber fractal]

[olive sable]

https://tenor.com/view/cat-generator-gif-22648083

https://tenor.com/view/cat-stealer-gif-21321506

[sage crag]

evilDentge

[rigid snow]

[mossy fossil]

Derailment?

[sage crag]

https://tenor.com/view/my-jolly-reaction-jolly-my-honest-reaction-gif-5490890317762380451

[mossy fossil]

🧐

[rough bloom]

ye real technique
run game, let DRM do its decryption/unpacking thing, then dump memory
you now have decrypted game

[sage crag]

wrr

[mossy fossil]

that actually is a genuine strat back in the C64 days

[sage crag]

dont deobfuscate code

[azure lynx]

just like analyzing malware.

[hard raptor]

Meow

[olive sable]

Meow

[rough bloom]

ye I don't, too annoying
it's one of the few things I rely on an LLM for
don't need it often enough to learn properly

[olive sable]

The duality of man.
How did i become a cat?

[rigid snow]

insert pieces of machine code server side along with other garbage that would crash the game on any different hwid

that it was requested from

not malware legitimate copyright law enforcement

[rough bloom]

[olive sable]

I should become a vtuber and then have that as my oshi mark

[rough bloom]

oshi lark

[olive sable]

Shh

Typo

Im on phone

[rigid snow]

tpyo

[olive sable]

No

Typo

[rigid snow]

tpyo

[rough bloom]

tpyo

[olive sable]

Happy newyear

[keen atlas]

happy new year!!

[sage crag]

birt

im fall asleep

thank you forever

meru new yewr

[rough bloom]

gn kni

[sage crag]

wrrrr

🍃

[ivory plinth]

Happy old year

[olive sable]

No problem

[quick condor]

My "simple project" is now loading piper, whisper large v3, 2x 24B models, 2x 8b models and an embedding model

[olive sable]

quality is terrible cuz file size limit

im also less likely to get doxxed this way i think

[tender river]

have an

evil new year

[olive sable]

[tender river]

me back to sleep

[fast pagoda]

i got an unsolicited invitation to a group for "dads with kids 0-18" on linkedin

which means i am now far too old

and must be killed

[olive sable]

oldge

[silent cloak]

I know those trees....

[olive sable]

damn

imagine knowing what a tree is

[fast pagoda]

muthafuckin bootleg fireworks

[azure lynx]

when does her memory consolidation happen? 😮
it seems like she might forget stuff that happens immediately before "going to sleep?" if there isn't a final pass by that layer before stopping her process.

[warped narwhal]

Too late. I've narrowed your location down to:

|| Earth (probably) ||

[olive sable]

[quick condor]

Well from my understanding there is probably the following:
Logs, chat transcripts, memories stored in some embedded format, and context.

In theory context will always be saved to transcripts but may not be saved to memories

Personally I do a fast embedding as part of the ongoing session on a rolling basis using a simple fast model and a deeper embedding after the session. Current session memories are pretty literal but memories are more broad and abstract. Forgetting is also important when you think about it especially for an Ai.

[warm river]

guys my xbox 360 broke 😭

[olive sable]

[warm river]

it turns off itself after 5 mins with red dot on power button

should i fix it and rgh it?

[olive sable]

is it a launch model or a sim?

[fast pagoda]

sounds like overheating if it works initially

[olive sable]

most likely overheating ye

[warm river]

xbox 360 S 2012 model

[olive sable]

otherwise psu maybe

[warm river]

it works normally for 5 minutes

then turns off

and power button has red dot and after restart it works again for 5 mins

[warm river]

red dot doesnt blink or flash its just static, and also i have to turn off it manually after red dot appears on power button

[olive sable]

the S solved the red ring

but they have the red dot instead

[warm river]

i mean console its not fully dead yet

[olive sable]

is it flashing red dot or solid?

[warm river]

solid

[olive sable]

o7

flashing is overheating, solid is ded

[warm river]

on screen it looks smth like that that it does glitch screen for second and becomes black

[warm river]

but it turns on?

[olive sable]

ye that sounds liek the gpu died

[warm river]

why it works for 5 minutes then

or longer

depending on load

[olive sable]

thermal expansion

once it gets hot the gpu disconnects from the motherboard

and then the system shuts off

[warm river]

then why after restart it starts working again

[olive sable]

cuz it got cold again

thermal expansion

[warm river]

that fast?

[olive sable]

ye

[quick condor]

I feel Vedal's "my heating bill isn't the problem, it's the electricity bill that's through the roof"

[nocturne olive]

The power of a heatsink with 0W going into it is quite impressive

[warm river]

it spins but its not that loud

[nocturne olive]

Meanwhile my GPU when any videogame: vacuum cleaner

The 3090 below the 4070Ti kinda limits the air the 4070Ti gets

[olive sable]

Does it look like this?

cuz if so that gpu is completely and utterly fucked

[warm river]

no

nothing similar to this

[olive sable]

then how does it glitch

[warm river]

just screen gets black in half for second

and it goes full black

[olive sable]

hmmm

if its only half the screen

[warm river]

its half only for one second

[olive sable]

it has to still be the gpu

[warm river]

it gets whole black after it

[olive sable]

jsut glitching differently

[warm river]

maybe gpu overheats?

i never cleaned that console

since i got in 2023

[umbral thorn]

shorted

solder point broke

[olive sable]

if it was overheating the light would be flashing. the consoleeither doesnt realize its overeating or the gpu is fucked

[warm river]

also it started happening randomly, i played with my cousin for hour today cod mw3 it worked fine then and when i booted pvz gw1 and after minute of running pvz it started doing that

later with just pvz started it started doing that non stop after 5 mins of running

[azure lynx]

if it ran for longer the first time, that sounds like overheating. if you let it sit for like an hour and then start again does it run for more than 5 mins the first time?

[olive sable]

get the error code form it.
turn the console on, press the sync button and keep it held down while you prress the eject button 4 times

then count the flashes

that gives the error code

[warm river]

no

but instead it freezes sometimes for minute

the whole screen

no red dot

[azure lynx]

that sounds bad. ;/

[olive sable]

☝️

[warm river]

i took console apart to clean it rn

[olive sable]

ah

[warm river]

idk if its good idea to turn on it while i took it apart

[umbral thorn]

console user

[olive sable]

nah the 360 is cool

[umbral thorn]

[warm river]

i play only on xbox 360 to play old cod

or pvz gw1

i planned to rgh it aswell

[olive sable]

probably not. but im afraid cleaning it wont help too much if its the gpu that needs to get reballed

[olive sable]

this is how my 360 kicked the bucket 2 years ago

[umbral thorn]

[warm river]

also the fan is pushing out a very very hot air in startup

[quick condor]

Yah I had to go to liquid cooling because cramming two fe 3090tis and an A4000 made airflow a struggle

[umbral thorn]

[warm river]

when i start it its already hot as shit

[umbral thorn]

wtf is airflow

[nocturne olive]

Liquid cooling too expensive

[umbral thorn]

[nocturne olive]

I guess allowing the GPUs to breathe

[olive sable]

it might just be shutting of from heat protection then

[umbral thorn]

wait me need find pc pic

[olive sable]

did you have the fan hole covered mayhaps?

[warm river]

shouldnt it blink then? or flash

[nocturne olive]

What do you mean not too expensive?

[olive sable]

the 360S already doesn't have too good cooling

[umbral thorn]

skill issue

[quick condor]

I mean it wasn't that bad, the A4000 block was like $110, the 3090ti blocks were like $150

[umbral thorn]

wtf expensive

skill issue

[warm river]

nop i placed it on flat table with position in vertical way

nothing blocks it

[olive sable]

you realize your definitions for expensive are a lot higher than ours'

[quick condor]

Compared to the rest of the system

[nocturne olive]

And where would I even put two radiators? Not to mention I have never waterblogged a GPU so I have no clue how to do it correctly

[warm river]

i placed it like its placed on xbox 360 box

[olive sable]

then it should have been fine

was there a lot of dust inside?

[warm river]

eh not that much

[olive sable]

fans failing?

[warm river]

i never changed thermal paste, either previous owner

[nocturne olive]

And where would I find cheap waterblocks for my 3090 and 4070Ti? And would that even be worth it?

[umbral thorn]

fckin rich peep

[warm river]

fan spin since i feel hot air when i start xbox