#programming

1 messages ยท Page 496 of 1

sage crag
#

like this

#

desmodder say it will capture like this, but it actually captures a full screen and then scales it because the viewport is locked

#

so i had to add an inverse transform to the

#

viewport

#

oh graphic didnt update

#

shame

#

i was hoping it would update to this

tender river
#

i decided to try manual scaling

#

its 1000000

#

no wonder f32 is having issues

#

wrr

#

it doesnt support f64

#

and turns out floating point is challenging here because of c

sage crag
tender river
#

NOT because of c

#

because of

#

c

sage crag
#

my fractal scales up to this zoom factor before reaching precision issue

#

or 81000x zoom

rigid snow
#

sdfkljsdlfkj

sage crag
#

fractal

rigid snow
#

i got a credit card

sage crag
rigid snow
#

eternal debt here i come

sage crag
#

what are the digits

#

:UNFOCUS:

rigid snow
sage crag
tender river
#

i can neuroPogHD

#

give them to me instead

rigid snow
sage crag
#

ye

#

i could sell them also

rigid snow
tender river
#

FOCUS gradients hard

rigid snow
#

dither

sage crag
#

or apparently not quite

#

close enough

tender river
sage crag
#

wrr

#

pixel

#

still using escape time colouring and not pixel colouring NeurOhISee

#

when rendering fractal you can do either

#

escape time or domain colouring

prime delta
#

Keep programming the universe or something...

sage crag
#

domain shows the complex derivative of the function at the point

#

and

#

escape time uses the escape time

#

but you can smooth it so its not stepped

#

this was the algorithm i used for smooth escape time colouring

#

math.ln(x * x + y * y) / 2 equivalent to ln(|z|)

#
log_zn = ln(|z|)
nu     = ln(log_zn / ln(2)) / ln(2)
smooth = iters + 1 - nu
frac   = smooth - int(smooth)

c0 = palette[k % palette.len]
c1 = palette[(k + 1) % pallete.len]

c = lerp(c0, c1, frac)

i guess

tender river
#

i did
iter -= log(log(log(zx * zx + zy * zy)));

sage crag
#

oh, nu could be rewritten as log2(log2(|z|)

tender river
#

log(log wasnt smooth enough

sage crag
#

hm

#

just thinking about it

#

do iterations of the escape time of the julia set increase in powers of 2

#

maybe thats why i used log2

#

so it could be cheaply computed on an int

#

but i forgot?

#

mayb

tender river
sage crag
tender river
#

i lied that wasnt the actual code

#

i translated it for brevity

#

but

#

code

sage crag
#
c0 = palette[k % palette.len]
c1 = palette[k++ % palette.len]

smooth = k - log2(log2(|z|))
c = lerp(c0, c1, smooth - int(smooth))
#

i guess smooth - k should also work

#

approximately

#

but they do exhibit different behaviours on a graph

#

so maybe not

tender river
#

should work well enough

sage crag
#

2.83kb

tender river
#

cybuero mostly boilerplate to support scaling

#

its my old code

#

the artifacts are really annoying

#

but this looks like actual fp artifacts rather than logic error

#

you can pan around with arrow keys

#

and scroll with scroll wheel

#

NeurOhISee the triangles sometimes get different artifacts

#

i wonder how easy it would be to fix this

mighty thorn
#

What version of Python do you all use

#

Ive arbitrarily chosen 3.10 as my main despite not knowing any of the differences and it being harder to make work for some stuff

sage crag
#

the

#

translation of original ableos mandelbrot

#

slightly strange looking

tender river
sage crag
#

using division by 32 is much clearer

tender river
#

always the

#

lines

sage crag
#

you can convert from mandelbrot to julia like this

sage crag
tender river
#

probably
-0.7269
+0.1889

#

oh i have an idea

#

i'll try same code but in

#

native webgpu

sage crag
#

colours

tender river
#

what function do you use for colors vedalMagnify

sage crag
#

random

tender river
mighty thorn
tender river
#

not fun

mighty thorn
#

WebGPU works (with great difficulty and convincing) in discord activities

tender river
#

i guess electron has no reason to disable it

mighty thorn
#

I got Qwen 2.5 0.5b to run entirely in the activity, and I double checked by killing the server script during inference and it still worked once all setup for the client

autumn kestrel
#

fractal neuro pls

tender river
# sage crag random

i think artifacts are related to how webgpu does things FOCUS native webgpu does similar things, but i tried similar code with opengl and it worked fine on large zoom

#

it could be wgpu in particular too
or some flags i need to pass
or something

stark needle
#

wrr

sage crag
# tender river i think artifacts are related to how webgpu does things <:FOCUS:1168267148523737...
vec3 hsv2rgb(vec3 c) {
    vec3 rgb = clamp(abs(mod(c.x*6.0 + vec3(0.0,4.0,2.0), 6.0) - 3.0) - 1.0, 0.0, 1.0);
    return c.z * mix(vec3(1.0), rgb, c.y);
}

void mainImage(out vec4 fragColour, in vec2 fragCoord) {
    vec2 uv = (fragCoord - 0.5 * iResolution.xy) / iResolution.y;
    
    //vec2 c = uv * 2.0;
    //vec2 z = vec2(0.0);
    
    vec2 z = uv*2.5;
    //vec2 c = vec2(-0.7269, +0.1889);
    
    const vec2 c = vec2(-0.65, 0.0);
    
    int i = 0;
    const int maxIters = 300;
    
    while (i < maxIters && dot(z, z) <= 4.0) {
        z = vec2(z.x*z.x - z.y*z.y, 2.0* z.x*z.y) + c;
        i++;
    }
    
    float s = float(i+2) - log2(log2(dot(z, z)));

    const float phi = 0.6180339887498948;
    //const float h = 213.0 / 255.0;
    float h = iTime * 0.2;
    const float hv = 0.1;
    
    vec3 col = vec3(
        //h + 255.0 * mod(hv * phi * sqrt(t) / s, 1.0),
        h + hv * sin(s * 0.2),
        //1.0 - 1.0 * sqrt(t),
        smoothstep(0.0, 1.0, s),
        //1.0 - 0.03 * s
        smoothstep(0.0, 0.9, s)
    );

    if (i == maxIters) { fragColour = vec4(hsv2rgb(vec3(h, 0.3, 1.0)), 0.0); } else
    { fragColour = vec4(hsv2rgb(col), 1.0); }
}
#

try

#

some cool

#

precision issue

tender river
#

gay

true hemlock
#

i added more iterations lmao

#

@sage crag

sage crag
#

how many?

true hemlock
#

as many as desmos could handle

#

before it just refuses to render

sage crag
#

you should be able to see it here:

true hemlock
sage crag
#

shadertoy

true hemlock
silent cloak
#

still crazy to me that u can do audio stuff through shaders

kind fable
silent cloak
#

kek

rough bloom
silent cloak
#

what even is it

sage crag
rigid snow
#

pytorch lightning is cool

sage crag
#

its an arc of elecricity that

rigid snow
#

blabhblahblahbla

#

nerd

fickle rain
rigid snow
#

them using claude is not a secret

fickle rain
#

Maybe we should have CLAUDE.md in Linux kernel/coreutils too

jagged turtle
true hemlock
#

in need of some opinion

#

of the IMU LSM6DSR

#

as far as i read it has magnetometer similar to other high end IMUs

#

just well slightly more prone to a bit of drift

opaque wharf
#

I think the LSM6 is accelero and gyro

#

Not magneto

thick karma
#

I'm finding some interesting shit inside osu's source code

#

presumably website source

stark needle
opaque wharf
#

Anyhow every gyro will drift over time anyway

thick karma
true hemlock
#

misread a datasheet

thick karma
opaque wharf
#

If you're filtering it anyway you can always try to correct the attitude of whatever you're measuring. But I think the LSM is more sensitive to temperature

true hemlock
#

i've built 4 tracking modules with LSM6DSR and was wondering if there's even more options i could try

thick karma
true hemlock
#

i think i'd need at least 5 or 6

opaque wharf
#

Or MPU6050

#

Other includes witmotion that provides AiO ready solution

true hemlock
#

2 on my torso for center reference

opaque wharf
#

If you're not low on budget I suggest just use witmotion. They're already packaged and have the options to use RS485 bus

true hemlock
true hemlock
#

so i build my own

opaque wharf
opaque wharf
true hemlock
#

i have one and the motion sensitivity is somewhat ass

opaque wharf
#

9250 is 6050 just with magneto inside the silicone

true hemlock
#

which supposedly makes it far better

#

though

#

my place doesn't particularly have stable magnetic environment

opaque wharf
#

I don't know about that. They are better because they have invensense proprietary filtering algorithm that can output quaternion

true hemlock
opaque wharf
#

Also, I think the witmotion does that too (outputting quaternion)

#

Invensense called it DMP IIRC

#

Digital Motion Processing

true hemlock
#

i could try witmotion honestly but for like maybe only one or two modules

#

they're expensive if i do like 12+ modules MONKAS

opaque wharf
#

How sensitive do you need it anyway and what's your bandwidth need?

opaque wharf
true hemlock
#

not much bandwidth, but definitely high sensitivity

stark needle
opaque wharf
#

Like blowing up the budget for government contract for example enub

true hemlock
#

which as far as i could find its mostly the LSM6 models that do well in non stable magnetic environments

true hemlock
opaque wharf
true hemlock
#

from the IMU that i have

stark needle
true hemlock
#

ah

opaque wharf
#

That replaces their old MPU model

true hemlock
opaque wharf
#

I told you both are discontinued enub

true hemlock
#

smh

#

the ICM models though

opaque wharf
#

Yes, same manufacturer

#

Invensense

true hemlock
#

oh, doesn't use magneto

opaque wharf
#

Because you don't need it enub

#

Unless you need absolute attitude

#

Like airplane

#

And they are a PITA to calibrate

true hemlock
#

its just trackers for vr anyway lmao

opaque wharf
#

If you even so much move an iron bar all bets are off

true hemlock
#

i specifically wanted to avoid those mainstream trackers like vive or others

sick owl
#

@fast pagoda La creatura arrived

opaque wharf
#

Comment: Finding legitimate MPU9250s has become exceedingly difficult due to counterfeits and DOA IMUs. Buy at your own risk.

stark needle
true hemlock
opaque wharf
#

You still need to filter them anyway enub

stark needle
#

Ok fair enough

true hemlock
#

i only have 8 LSM6DSR though

#

and turns out it was one of the best option

#

oh well

#

its fine i can just

#

order more for like $2 each

stark needle
true hemlock
#

okay so

#

im on the right track

opaque wharf
#

My suggestion is use STM32 rather than ESP32 for the processing and filtering

true hemlock
#

if i could pull this off i might have something of a quality of slime trackers

opaque wharf
#

Use the ESP purely for the wireless stuff

stark needle
opaque wharf
#

And other high level stuff

stark needle
#

Like obviously it's not gonna beat lighthouse trackers

#

But like lighthouse is it's own pain anyway cause if u have anything in ur room or want to put a blanket they're dead

true hemlock
#

i'd want it to be able to connect to my network wirelessly

#

probably wifi

opaque wharf
#

Words of warning, prepare for latency enub

true hemlock
#

i don't like trackers that can be obstructed

#

and they're

#

overpriced

#

ngl

stark needle
#

That's what i did

true hemlock
#

i have my own wifi access point in my room

#

connected to my switch

stark needle
#

Yeah then ur good

opaque wharf
#

Yeah that's fine then

#

Will not be as bad

stark needle
#

U can get like ~50-80ms latency

true hemlock
#

its how i can comfortably also play wireless pcvr

opaque wharf
#

If you want even lower then ditch the esp entirely and use dedicated 2.4GHz IC like some nrf offering

true hemlock
#

fair

stark needle
#

Yeah but gl diying ur own solution like that

#

The problem is the drivers

true hemlock
#

i study ic datasheets and pinout for life

opaque wharf
true hemlock
opaque wharf
#

What is the driver usually for vr tracker anyway? Not HID right?

#

If it's HID then it's easy enough

stark needle
#

wait what's ur vr headset @true hemlock ? i suppose pico or quest?

true hemlock
#

quest 3s

#

128gb

#

i don't really feel the need to spend more than that

#

i think i got it for $250

#

new

stark needle
#

not worth to dump 1k on a headset when most vr looks crap anyway

true hemlock
#

^

#

and i don't need more storage

#

i do pcvr anyway

stark needle
#

i mean if u just dont want lighthouse u can just buy 3 vive ultimate trackers which are still cheaper than lighthouse+vive trackers

true hemlock
#

uh

stark needle
#

but theyre like 600$ in total

true hemlock
#

yeah... nah

stark needle
true hemlock
#

i move around few rooms

stark needle
#

yeah fair enough then no

true hemlock
#
  • sometimes curling into blankets are fun
#
  • i like good battery life modules and lightweighted
#

so i figured

#

i can just

#

maybe spend extra $30

#
  • the stuff i already have
#

and build a full set of trackers

true hemlock
#

how much do they cost

stark needle
#

like 380 euro here

#

for all

true hemlock
stark needle
#

and supposedly u can also use them for mocap

true hemlock
#

oh

stark needle
#

ok no supposedly they're BMI270 accordinf to some reverse engineer

#

or ICM40608

true hemlock
#

nvm then

#

basically more pricy slimes

stark needle
#

yeah

#

just get slime atp

#

with the trackers u alr have

#

u need like 5 trackers only

opaque wharf
stark needle
opaque wharf
#

I don't know much about tracker but the fact that the update rate, the zeroing, and all that jazz is shitty compared to what's actually need for smooth experience

stark needle
#

has direct 2.4ghz as well via dongle

opaque wharf
#

Oh

#

I read that as shitfall at first evilWheeze

#

Like, who tf names their company or is that an open source project. But the site design screams company evilWheeze

opaque wharf
#

Ye, shift all not shit fall enub

stark needle
#

lmao their manual is a google doc

opaque wharf
#

That's one way to do it I guess. Faster time to iterate than updating the site

stark needle
#

@true hemlock supposedly these even have lidar

opaque wharf
#

I would say that lidar will not be uhhh, as effective as they made it out to be

stark needle
#

better than nothing

opaque wharf
#

Yeah

#

But they also have the ankle version

stark needle
#

also supposedly also works in standalone

#

via phone

#

and they have a pro version for better ankle/less cables

#

for 398 euro

#

omg wtf

#

check out what else they sell

#

"soundproof microphone"๐Ÿ’€

true hemlock
#

what the actual fuck

stark needle
#

is this what peak vr looks like

#

all just to be an anime girl online

true hemlock
#

bruh looks kidnapped

rough bloom
#

ye he is

#

they didn't pay him for this ad

true hemlock
stark needle
#

imagine ur parents walking in on u like this

opaque wharf
stark needle
#

and theyll send u to the psych ward

idle dune
stark needle
#

lmao

#

imagine showing up like this to ur boss

#

hi it

mighty thorn
mighty thorn
#

I want to see it is typing

vague sonnet
stark needle
#

it is typing

mighty thorn
#

Suzukiโ€™s did u u n g h h n idk did did did did id did id do did six fix. I dis six six so. Six er f c ff a s ff d d. CJ. J I kid d d d gg d s s doc I jd do I j j b g t to wi he sdjdhddhddjdjdjdjdsiizxjsj s dedication ii j j inch u j n b kids didndjdjdjic j i j odd do. Idiocy u y labs Sidney u Judd CJ n ex in ex in ex in ex in on ex to d into s hi k de h c n b in l s c

#

Alrighty thanks

mighty thorn
#

Am running tests of my prototype conditional depth model

#

Itโ€™s gotten up to 82% accuracy so far

#

Pretty poggies

true hemlock
#

bruh

kind fable
#

i was slow for this one

#
mighty thorn
true hemlock
kind fable
stark needle
#

ive been coding for 5-6 years and i still only know 2 regexes

#

* and **

kind fable
mighty thorn
kind fable
#

that's globbing maybe but not regex

stark needle
#

bruh

#

then i know 0 regex

opaque wharf
#

That is a valid regex pattern to match for any character

kind fable
#

nope that's a glob

opaque wharf
#

*+ will select everything

mighty thorn
# stark needle your what

L>L>L>L>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
/
IL>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>L>OL

kind fable
#

a regex to match any character once is .

opaque wharf
#

* will select just one char

mighty thorn
#

Thatโ€™s probably formatted correctly on desktop

#

Probably

kind fable
stark needle
#

whats a conditional depth model

true hemlock
#

fuck lmao

mighty thorn
kind fable
true hemlock
#

that's my first attempt ig

mighty thorn
#

Well the Lโ€™s are layers

true hemlock
#

idk how to read regex i do it by vibes

mighty thorn
#

The IL and OL are input layer and output layer

stark needle
kind fable
#

it's not even like the regex here is hard

opaque wharf
#

Ohh right

true hemlock
#

i know its why i could solve it

kind fable
#

all you got is OR and wildcards here

opaque wharf
#

* is 0 or more

#

I forgot

true hemlock
#

i don't actually really know how to read actual regex

kind fable
true hemlock
#

this one is pretty self explanatory

kind fable
#

in regex if you want to say any char you use a dot

opaque wharf
kind fable
#

and then you use the star modifier on it

opaque wharf
#

You can't just put star

#

You need to escape it

kind fable
opaque wharf
#

a* means 0 or more a

true hemlock
#

this one i assumed "must contain vt" along with other random shit i just solved by looking at the other regexes

kind fable
opaque wharf
#

Hence you can't just type * to regex engine as that would cause it to modify something

tender river
opaque wharf
#

You need to type \*

kind fable
rough bloom
true hemlock
#

considering the reaction

rough bloom
#

short and always the same format

true hemlock
#

i assume my time is pretty bad

kind fable
#

i know for sure one bad engine which allow not escaping the modifier if it's the first character

true hemlock
kind fable
#

i forgor which one tho

#

but yeah escape stuff

true hemlock
#

also now y'all caught me never used any regex before

kind fable
#

anyway

#

REGEX is so simple and easy

#

everyone should know regex

true hemlock
#

fuck regex

stark needle
#

damn kaine is writing an essay

true hemlock
kind fable
#

regex is perfect to validate URLs
just do .+@.+ smh /hj

true hemlock
#

better than my astrophysics rant

stark needle
#

lmfao

true hemlock
mighty thorn
# stark needle so just residual/hyperconnections?

Basically do all of the input tokenization, preprocessing, multimodal embedding, etc etc using the input layer, then let it flow through the first few weight layers (the tail), then cache the outputs of everything, route the current activations and shit of the final tail layer to a secondary set of layers (the reader layer stack) which predicts a single next token (the string within <|role|>โ€ฆ<|role|>), fast forwards up to the output layer, then if the predicted next turn is of the assistant/model role then you reload the cached tail state and let everything flow through the rest of the main model layers (the writer layer stack) and generate an output normally, but if the reader predicts that the next turn wonโ€™t be an assistant turn then you stop there.
Is completely useless on virtually all pretrained models since they almost 100% all use just two real roles, โ€œhumanโ€ (user) and ai (model/assistant), but my future plans entail a model with a more open ended role system where, if these tests validate the core concept, this can be used to solve one of the biggest expected issues Iโ€™d otherwise have

true hemlock
#

i got curious on theoretical scenario where you have near endless supply of energy or just higher than C^2:1 energy to mass ratio conversion and storage

true hemlock
#

figured that if you could just accelerate at 1.5g half the time and decelerate on the other half

#

it takes about < 20 years for you. probably about 18 years

#

takes millions of years from earth's perspective though

#

~2.5 million years

#

special relativity is both a blessing and a curse

stark needle
# mighty thorn Basically do all of the input tokenization, preprocessing, multimodal embedding,...

so just early exit? came out in 2024 (https://arxiv.org/abs/2404.16710) and was integrated into HF transformers ages ago https://huggingface.co/blog/layerskip

Weโ€™re on a journey to advance and democratize artificial intelligence through open source and open science.

mighty thorn
#

But that same concept

#

And also the fact that itโ€™s more of a railway junction than an early exit cause said exit has its own layers and outputs

#

The tail stays frozen tho

stark needle
#

like

#

wym predict the turn

#

im confused

#

whats the purpose of this early exit

mighty thorn
#

The bold is the text it is generating

#

The identity of the next turn, before it is actually sent

stark needle
#

so like you want to speculatively find out when a person is about to talk?

#

im confused

#

no one lets their llm generate the role tokens

true hemlock
#

im confused aswell

#

so im using a generalized term

mighty thorn
#

The next message after an assistant turn will always either not exist or be a user turn

#

But if there are more than 2 roles

#

And you only want a response generated for 1 of them

stark needle
mighty thorn
true hemlock
#

so you wanted the model to predict if the next message will be from another user

#

and you wanted the model to predict when it can start responding as in no one else is going to talk next

mighty thorn
#

I want the model to use the multi user group chats it will be trained on, and the natural turn taking dynamics learned from them, to choose when it talks as if it were just sitting on its keyboard reading chat and waiting

stark needle
#

this is just an inference harness problem

mighty thorn
#

People randomly join servers

#

AFK in middle of convo

#

Etc

#

And multiple servers too

true hemlock
#

or you can just loop a normal llm with function calling and let it decide whether to respond or wait with some prompting

stark needle
#

true

mighty thorn
#

Yes, but this is more efficient
Plus asking an LLM to only speak when appropriate is like asking a toddler to only cry when they are scared

true hemlock
stark needle
#

just need a reward function

vague sonnet
true hemlock
#

i wanna drop a copypasta

stark needle
true hemlock
#

but i'll need to predict if i'd get smitten by Zeppelin or not

stark needle
#

๐Ÿ’€

true hemlock
#

with my secondary line of neuron propagation

stark needle
#

drop in my dms i wanna see it lmao

mighty thorn
stark needle
vague sonnet
true hemlock
mighty thorn
#

โ€œSNN cheaperโ€
look inside
$50000 entry level biomorphic hardware that doesnโ€™t exist for models that donโ€™t exist

vague sonnet
#

sounds like you don't know how it works

#

but that's fine

opaque sigil
stark needle
mighty thorn
# stark needle grpo

Yeah well I could also just use OpenAI api for the entire project and run it off a raspberry pi but thatโ€™s no fun is it?

stark needle
#

qwen 3.5 uses basically no kv cache

mighty thorn
stark needle
#

qwen 3.5 uses gated deltanet

#

for like 80% of layers

mighty thorn
#

Bro why try experiments and learning from doing them when you could just consume and settle for a worse option that isnโ€™t yours

#

Ur so right

#

(this man has stock in Alibaba)

stark needle
mighty thorn
mighty thorn
# stark needle problem solved (?)

You know if you had an idea that you turned into a concept and made a prototype of and were really excited about, i wouldnโ€™t be the one to go โ€œerm actually this already exists but in a different less interesting way in conclusion stop being happyโ€, i would ask for details then give suggestions on potential modifications and/or applications you hadnโ€™t thought of

true hemlock
#

welcome to the scientific fields

stark needle
#

with less kv cache usage

#

than a standard transformer

#

u can always modify it

mighty thorn
true hemlock
#

ya see

#

end user usecases aren't usually solved with complete architectural modifications

#

what you wanted to do was done many times by people using their own methods of calling the inference

#

shapes.inc bots are one example

#

plenty of other backend tricks

mighty thorn
#

Iโ€™m still gonna do it

#

Cause I think itโ€™s cool

#

And also 6/8 tests are done now

true hemlock
#

im not stopping you, nor im trying to

#

its a cool experimentation

#

but im just conveying what shadow is trying to say

mighty thorn
#

I know

#

Thank you tho

#

Wasnโ€™t mad at him or anything SMILE

true hemlock
#

i personally am not trying to downplay you. honestly i think its cool that you're thinking of your own ideas

sage crag
#

wrr

mighty thorn
true hemlock
#

clumsy experimentations is what shaped the scientific community after all, and no experimentation are non clumsy

opaque wharf
#

Immortal venerable is here enub

#

Bow down you guys enub

sage crag
#

i am no longer immortal or venerable

opaque wharf
#

Lop enub

#

Ying xor Ying is null enub

#

Null or lop, lop it is enub

mighty thorn
true hemlock
#

nop

tender river
true hemlock
#

i love spamming nop in my ASM

opaque wharf
#

My cats reaction to that information

#

He's eepy

#

Luckily I managed to move him before I held hostage enub

opaque wharf
#

Apparently 3 days worth of Nitro is worth 15 minutes of your attention span enub

silent cloak
#

(which already makes it overly easy)

kind fable
#

but don't tell discord

opaque wharf
#

I'm calling John Discord right now enub

kind fable
#

noooo

umbral wigeon
#

๐Ÿ™ ๐Ÿ™ ๐Ÿ™

#

๐Ÿชฃ ๐Ÿชฃ ๐Ÿชฃ

stark needle
#

U define a trainer

#

Then do .train()

#

Or similar

#

And u work with callbacks and provided loss functions and stuff

#

Instead of writing ur own torch training loop

#

And it handles parallelism and annoying accelerators like TPUs

sage crag
#

neurOMEGALUL the compression is so bad

#

@tender river the

#

thing

tender river
sage crag
#

moire vs jpeg

#

jpeg win

#

i find it interesting how it appears to get linearly bigger, but its actually scaling exponentially

#

after the 13 seconds are up its already scaled 440000 times

simple sky
true hemlock
sage crag
#
vec3 hsv2rgb(vec3 c) {
    vec3 rgb = clamp(abs(mod(c.x*6.0 + vec3(0.0, 4.0, 2.0), 6.0) - 3.0) - 1.0, 0.0, 1.0);
    return c.z * mix(vec3(1.0), rgb, c.y);
}

void mainImage(out vec4 fragColor, in vec2 fragCoord) {
    vec2 uv = (fragCoord - 0.5*iResolution.xy) / iResolution.y;

    const vec2 p = vec2(0.75621304, 0.38042766);
    float s    = 2.0/exp(iTime+0.0);

    vec2 z = p + uv*s* 1.15;
    vec2 c = vec2(-0.8, 0.0);
    
    //vec2 z = vec2(0.0, 0.0);
    //vec2 c = p + uv*s* 1.15;

    const int maxIters = 43;

    const float phi = 0.6180339887498948;
    const float h   = 170.0 / 360.0;
    const float v   = 0.02;

    vec3 col = vec3(1.0);

    for (int i = 0; i < maxIters; ++i) {
        z = vec2(z.x*z.x - z.y*z.y, 2.0*z.x*z.y) + c;

        if (z.x*z.y > 1.0) {
            float t = float(i+1);
            float k = t/float(maxIters);

            col = hsv2rgb(vec3(
                h + v*phi*t,
                1.0 - 0.9*sqrt(k),
                pow(1.0 - sqrt(1.0-k), 0.6)
            ));
        }
    }

    fragColor = vec4(col, 1.0);
}

put that in shadertoy if you want to render it

#

swap the z and c stuff to switch between mandelbrot and julia

#

p is the centrepoint of the render and s is the scale

olive sable
sage crag
#

feel free to

#

add msaa

#

the

#

shader

#

is

#

right

#

there

olive sable
#

i

#

might

#

later

#

doing

#

unity

olive sable
#

i ownt

kindred perch
#

helu

olive sable
#

elo

kindred perch
#

sup sam

olive sable
#

hi

opaque wharf
#

hi sam

frigid trail
#

hallo

opaque wharf
#

I am now having NOยฒ enub

sage crag
#

@tender river enub

thorny marsh
sage crag
#

ye

stark needle
#

(ying xor ying) or lop

thorny marsh
#

hows yall doin programming peeps

opaque wharf
olive sable
#

bwaadow could be better

mighty thorn
thorny marsh
mighty thorn
sage crag
#

1m view

thorny marsh
silent cloak
#

Thems are fighting words around here

warped narwhal
mighty thorn
#

Yeah itโ€™s like that

warped narwhal
#

I think therefor I am argument Minamhm

sage crag
#

i am argument

silent cloak
#

Internet users

sage crag
#
int main(int argc (<- me), char **argv (<- not me)) {
  return 0;
}
opaque wharf
#
int main(int a, char **b) {
  return 0;
}
silent cloak
#

What is this and why did u do it

opaque wharf
#

Is that still you?

sage crag
#

only exists in

#

c89

#

or maybe c88

kind fable
#

nope always exist

sage crag
opaque wharf
#

Alright so this is you?

int sum(int argc, int b) {
  return argc+b;
}
silent cloak
#

Keep in mind C allows up to 12 levels of pointers

#
int ************ptr
sage crag
opaque wharf
sage crag
#

look theyre right here:

void *p
umbral wigeon
#

https://copy.fail

Reduce your response time to credential compromise with Flare @ https://go.lowlevel.tv/flare2026

๐Ÿซ MY COURSES
Sign-up for my FREE 3-Day C Course: https://lowlevel.academy

๐Ÿง™โ€โ™‚๏ธ HACK YOUR CAREER
Wanna learn to hack? Join my new CTF platform: https://stacksmash.io

โŒจ๏ธ KEYBOARD
Like what you hear? Grab a Q5 at ht...

โ–ถ Play video
silent cloak
#

ANSI C I mean

kind fable
sage crag
opaque wharf
kind fable
#

also no in ANSI C23 it's actually 275 level of pointers i think

opaque wharf
#

Or was that later addition?

kind fable
#

and it's a bottom

tender river
#
void *t
sage crag
#

hi t

stray dragon
true hemlock
tender river
#

hi t

opaque wharf
#

Hi

int sum(int argc, int b) {
  return argc+b;
}
sage crag
silent cloak
#

3 star programmers are something else

opaque wharf
stray dragon
opaque wharf
#
int sum(int argc, int b) {
  return argc+b;
}
stray dragon
sage crag
#

hi t

stray dragon
sage crag
#

hi t

stray dragon
stray dragon
umbral wigeon
#

-1

silent cloak
#
void * (*(*fp1)(int))[10];
#

This is truly an abomination

kind fable
silent cloak
#

But ive seen it before

#

All the best C books are insanely overpriced

umbral wigeon
#

Can I run holyC in window or linux?

#

I'm wondering

silent cloak
#

No the OS would burn from its righteous power

umbral wigeon
#

It would be funny if HolyC works with gcc and modern build system

silent cloak
#

Ive always meant to actually look into how it works

kind fable
silent cloak
#

I doubt its using LLVM

kind fable
#

wait C29 is awesome i'm gonna use that

Add the defer keyword and defer blocks and statements. This is similar to Go or Zig which both use defer for cleanup. Compilers like GCC and Clang have implemented this specification.
no more needing to use goto error pattern

opaque sigil
silent cloak
#

Are they actually going ahead with one of them finally?

kind fable
#

dam this is also amazing, scoped stuff like i can do in Java

For the if statement, add support for variable declaration, such as if (int err = xmit_ch('X')) log_err(err);, similar to how the for change was added in the C99 standard. The new variable can be used inside the scope of all branches of if, else, else if too. In C++, variable declaration has been allowed since the C++98 standard, further enhanced in the C++17 standard to allow for a second clause so the variable could be used within the condition itself, and C will use the later for C++ compatibility.

opaque sigil
#

iirc defer was supposed to make it into c23 originally?

silent cloak
#

Might be a nothing burger

#

People have been begging for a solution for decades and they usually shoot them down

opaque wharf
silent cloak
#

If they are actually going ahead and adding it then I'll listen

kind fable
#

daaam so cool

For statements that are iterations and for the switch statement, add naming via labels and transferring control through named break and continue (multi-level breaks). The syntax is the same as in Java.

silent cloak
#

And C#

opaque sigil
kind fable
silent cloak
#

So its actually in the specifications?

kind fable
kind fable
silent cloak
opaque sigil
#

oh right, it required lambdas enub

kind fable
opaque wharf
#

Ain't no way

silent cloak
silent cloak
#

No promises

kind fable
silent cloak
#

They do that with draft features sometimes

kind fable
silent cloak
#

Doesn't mean it will be in the final

#

its like the half life 3 of systems languages

opaque wharf
#

This is design by committee we were talking about so yeah. If some asshat doesn't agree it could be removed enub

silent cloak
#

Sadly wouldn't be the first time

opaque sigil
#

time for another extension FOCUS

silent cloak
#

C+-

opaque sigil
#

I should try this out though apparently it's in clang 22.1 NeurOhISee

silent cloak
#

C* (no memory safety)

kind fable
opaque wharf
kind fable
silent cloak
opaque sigil
#

surely you're not forced to use a whole block right

kind fable
#

i migth switch back to C from C++ just for defer
i only use C++ for the slightly bigger STD anyway

umbral wigeon
#

What modern C++ do you use

kind fable
#

C++23

opaque wharf
#

I wonder if wax seal can still be accepted for legal document enub

umbral wigeon
warped narwhal
obsidian mantle
#

wtf is this thonk
build stuck on this and nothing is happening

amber fractal
#

Needs VS to proceed

obsidian mantle
#

but its ...

#

vs's terminal

#

building through xmake is the same...

#

wtf

#

i guess its same thing

amber fractal
#

Jarvis, swap between C and Rust every 3 lines

silent cloak
#

still meaning to actually look into xmake

#

heard good things

kindred perch
olive sable
#

Jarvis make a C compiler with rust, and use it to make a rust compiler in C

kindred perch
#

Jarvis make AGI and make humanity extinct

amber fractal
#

Proud to sacrifice myself to the machine god

kindred perch
opaque wharf
#

Unless you're really into LUA

umbral wigeon
#

It's made in china

#

And I like lua

olive sable
#

jarvis delete xmake

opaque sigil
#

xmake is nice but i remember when i tried converting one of my libraries from meson to xmake it ended up being quite messy because vendoring dependencies alongside patches is i guess not something they really want you to do

umbral wigeon
#

Lua is easy bro trust me you will like it

kindred perch
#

Jarvis hold my beer

olive sable
#

Ran out of ram

#

Pc froze

opaque sigil
#

open different tty and kill whatever uses it all

opaque wharf
kindred perch
opaque wharf
mighty thorn
opaque wharf
#

But just the REI

kindred perch
mighty thorn
#

Has 256gb btw

mighty thorn
olive sable
#

Its RMA

#

Im on 2x8 rn

kindred perch
opaque wharf
#

The magic SysRq key is a key combination understood by the Linux kernel, which allows the user to perform various low-level commands regardless of the system's state. It is often used to recover from freezes, or to reboot a computer without corrupting the file system.
This key combination provides access to features for disaster recovery. In th...

opaque wharf
olive sable
#

16gb total ye

opaque wharf
olive sable
umbral wigeon
#

Xmake top 1 china build system

mighty thorn
olive sable
opaque wharf
amber fractal
#

Anti political bomb

umbral wigeon
#

It's 00:32 am

#

Let sleep

olive sable
#

For you

umbral wigeon
olive sable
#

Issue seems to be unity. Time to nuke

opaque wharf
sage crag
#

"issue seems to be unity" is not a classic

#

a classic is some fine wine

#

"issue seems to be unity" is tap water

#

its everywhere

olive sable
olive sable
#

Jarvis, delete your own existance im getting real tired of this meme

olive sable
#

hmm
i wonder if my desktop support LRDIMM

#

then i can go from 16gb to 256gb

#

newliv numlock

#

No, the AMD Ryzen 9 5950X does not support LRDIMM
meow

sage crag
#

my desktop supports HRAM

sage crag
#

it does not work well

olive sable
#

i seent

reef ingot
#

Day something of making my own game engine

Did nothing today

rough bloom
#

ECC UDIMM should work though

sage crag
wispy pike
opaque wharf
sage crag
#

highly contentious

olive sable
wispy pike
# kindred perch Disney is broke bro

what part of "so much worse" do you not get? Jarvis will get funding from the worst elements of the AI backer space because it would have absolutely no fitler nor morals

opaque wharf
#

A shame that they make ECC "enterprise" enub

rough bloom
olive sable
#

in server

opaque wharf
#

As John Linux famously said, he needs a machine with ECC enub

rough bloom
#

4x64 is an interesting choice

#

8x32 or something would've been more reasonable

olive sable
#

its cuz at 64gb you can fill the cpu's whole cpaacity

opaque wharf
olive sable
#

each cpu has 6 lanes and 12 rams lots, and supports up to 12*64gb of ram

sage crag
rough bloom
sage crag
#

lutel

#

pluu

rough bloom
opaque wharf
rough bloom
#

mabe

olive sable
#

mayhaps, i doubt it tho

opaque wharf
olive sable
#

it got marketed as 4x32, and when i opened it up itr was 4x64. so the guy didnt even know what he had

opaque sigil
#

i hate this language, tf you mean this took another couple years neuroCry

olive sable
#

but more like sales

#

and data

#

not hardware

opaque sigil
#

no, having println without any arguments for just a newline

sage crag
#

let me introduce

#

println("");

opaque wharf
#

They look at arduino and think "yeah, that's good"

#

Speaking of, I can't believe how many deployed hardware is being run using arduino right now

#

The framework not the hardware

opaque sigil
sage crag
#

print(line);

#

abstraction

opaque sigil
sage crag
#

#define line '\n'

#

feature added

opaque wharf
#

You can't do print("something"line); enub

#

Fix your define enub

olive sable
#

just dont print

opaque wharf
olive sable
#

printers suck

opaque wharf
#

Not by much on linux tho

#

I remember I made a printing shop system with membership card using raspi and nfc module

#

Good times when sbc are actually affordable and slim enub

sage crag
#

try running that and see what happens

#

i promise you it works

opaque wharf
sage crag
#

maybe

opaque wharf
#

I am on my bed rn enub

#

So can't check with c compiler enub

sage crag
#

im also on a bed and cannot check

opaque wharf
#

@olive sable

olive sable
#

no

#

unity

#

bwaawawawa

opaque wharf
#

@opaque sigil

olive sable
#

i am making player and camera controller

sage crag
#

@olive sable

olive sable
#

and its bad

sage crag
#

@olive sable

opaque sigil
#

hm?

olive sable
#

hi

sage crag
#

@olive sable

#

open godbolt

opaque sigil
#

you can't concatenate a string and a char

sage crag
#

r

#

whatever

#

bad language

olive sable
#

i dont have godot

#

or godbolt

sage crag
#

#define line "\n"

opaque sigil
#

that does work obviously

sage crag
#

godbolt is a website you stone

opaque wharf
opaque sigil
#

not in c++ without a space between tho

olive sable
#

what do i have to put into godbolt?

rough bloom
thorny marsh
kind fable
opaque wharf
sage crag
sage crag
#

i bedrot

olive sable
kind fable
sage crag
#

but actually im not bedrotting im sitting outside

opaque wharf
opaque sigil
#

your code is broken

#
#include <print>
#define line '\n'

// needs -std=c++23
int main() {
  std::println("bwaa"line);
  return 0;
}
thorny marsh
sage crag
#

I NOT BEDROT

sage crag
#

redundancy is good for runtime safety

opaque wharf
thorny marsh
sage crag
#

msvc

opaque wharf
sage crag
#

other compiler

#

mostly

#

clang

#

which supports gnu++ versions

#

if you only care about the cstandard you cant use it though

opaque wharf
#

But does is support all the extensions?