programming

[tender river]

its also about finding people, about many-to-many conversations, both public and private

and of course about storing all that

i mean i used to think p2p would be cool to use for messaging as well, but federation works already, while with p2p there are many unsolved challenges that plague networks like freenet

sorry its hyphanet now

[prime ridge]

no I get it

although filesharing should def be p2p

p2p messenging has a slim usecase

if I want to send a private message to one person and one person only then yeah p2p is better

although torrents are already a thing

idk I just find it cool that's kinda all lol

[tender river]

even then federation is preferred (imo) as it allows entrusting your data to a centralized provider rather than storing it yourself, for some it may be a downside but for most it wont be, and if its a downside you can run your own instance

no issue there as long as encryption is used

[prime ridge]

depends where you live too

there isnt always a federated server nearby

although I admit that kinda is just getting into the weeds

[tender river]

its text, you will live with 500ms latency

[prime ridge]

right

that's why I almost entirely use federated services

but I do prefer p2p filesharing

for large files

[tender river]

for file storage usenet is about the only federated approach

[prime ridge]

not storage but just transfer

[tender river]

same thing

[prime ridge]

not really 😂

well for p2p ig

[scarlet arch]

Guys on Sunday is Towel Day

[ivory tiger]

I hate coding stuff to support unstable Internet connections

[stark needle]

I hate coding stuff

[opaque sigil]

I hate

[tender river]

I love

(but also hate heheh)

[amber fractal]

I love hate code

sometimes

[hoary lion]

You definitely code

[amber fractal]

I'd not call one project in particular code, Not even god himself could attempt to explain why that codebase works.

[olive sable]

https://tenor.com/view/eroi-eroi-the-master-goku-black-goku-return-of-the-king-gif-3720166435547306034

Hi

Its 1 am

[amber fractal]

Go eep! @olive sable

[olive sable]

I gotta catch a plane to portugal so no

[amber fractal]

o7

[opaque sigil]

bored and decided to throw a cursed nvlabs library at jules, surely it can fix its non-existent include statements so i can actually work on it right

whoever at nvidia decided to solely rely on a bunch of makefiles to supply headers alongside using .h for what is really just a bunch of kernels, i hope your pillow will be warm on both sides tonight

[amber fractal]

I'm learning it is stupidly hard to both keep the bingo's shape while also getting all the elements to change size depending on the filled box.

Seems like I either get auto size or auto wrapping, can't have both

I'll probably go with auto size, wrapping isn't that large of an issue.

[prime ridge]

im ngl Jules is really bad at programming

they def didn't use the 2.5 pro model

it's probably 2.5 flash or sumthin

it reminds me of the GPT 3.5 days where it took 5 tries to create a simple 30 line script

and it's sooooooo slow and buggy

I only use it for the context window and often time I have to manually fix it's bugs anyway

[amber fractal]

Both approaches require javascript, very nice indeed.

I think then I'll ignore sizing for now, The goal was to see what I needed to do on the web server side.

[olive sable]

[amber fractal]

I still prefer the orginal over this mess

Doesn't even have the current progress smh

Ik the recommended apporach which I may actually do for bingo is just draw the lines in SVG, I'm not doing it in the FAQ as I don't want to include JS in that one yet

[midnight sigil]

imagine vibe coding

[amber fractal]

Couldn't be me

[olive sable]

[amber fractal]

I do wish that one was more exact because depending on what, it should be crossed off

If it is meant to be either "debating vibe coding" or "chat member asks to debug AI code", the first already happened today

[opaque sigil]

[midnight sigil]

I hope I can manipulate gpu kernel from geometry nodes-

well at least it tells me that geonodes are really utilizing my whole cpu

raycasting 980418560 rays is insanely costly

[opaque sigil]

i just wanted to look at how nvidia implemented bigints, why do i have to rewrite the entire library to get a sane editor experience

[amber fractal]

I love magic numbers

[hoary grove]

how about claude4 opus?

have someone tried？

[olive sable]

Whats with all these claude 4 things? Is it limited or something?

[prime ridge]

it's new

[hoary grove]

Claude new model

[sullen marsh]

sounds like neuro lol

Same question, different AI, same answer

[barren ginkgo]

So true

I got myself into that ai coding rabbit hole myself

Made my own model and when I said that I might go back to older version ( to test her reaction ) she begged me to spare her

[olive sable]

Each time a file finished transfering to my tablet the file transfer crashes and i have to reconnect

Its usb-c to usb-c and less than a gig of data

[opaque sigil]

thanks for reminding me i've been meaning to get a new usb data cable

[amber fractal]

Alright, magic numbers later and grid has been upgraded to decent

Now all that is left for today hopefully is overlaying a second grid containing the crossed out sections

I just need to figure out how I'm doing that

uhhhhhhhh

I don't know how I should do this

Best I got is background image of an X

-# One of these days I need to figure out how the other half of this project is going to work

[olive sable]

I got it to work

I got my movie on my tablet now

[old totem]

help
(i know i could just put the print inside the function)
but i wanna know how i can pass an input inside a function as an argument

im not good with pointers...

[old totem]

"I want to study AI" why do i feel like i just got hit

[amber fractal]

The goal was to hit common events, welcome

[amber fractal]

I should say this is a copy of the chart, I know the original needs to be updated

[hoary lion]

oops

got like third of the bingo pointed to me

[amber fractal]

I'll be listing extra context in my version but I know it is supposed to be topics of the channel

[hoary lion]

rare vedal appearance when

[olive sable]

Airplane mode time, going dark

[hard shale]

this results into unspecified behaviour, and plus you just did pass an input to a function. unless you meant pass by reference instead of value.

if opt for a pointy-approach
in randomGame's call: guessVal should be an address, for instance &guess
in randomGame's definition, guessVal should be a pointer to an integer-type. Therefore scanf(format, $guessVal) must be express as scanf(format, guessVal) to guarantee correctness.

otherwise with your following approach:
keep in mind that a function-type unlike other languages can only return one value, therefore return 0, 1 is invalid.
guess should be initialized with randomGame return-value, instead of omitting the values. That'll be equivalent of (void) randomGame(arguments); where the expression will be evaluated but the values discarded

look for a book online like The C Programming Language, 2nd Edition it will benefit you a lot.
but there's a catch! it's a super old book regarding C-ANSI and it will cover some things that shouldn't be a practice anymore.

[old totem]

scanf forcibly makes you input the address so if i remove & it ends the terminal

void doesnt let me return because idk why

thanks for help tho

[old totem]

i'll check it out!

[hard shale]

void is an incomplete-type, it doesn't fit the requirements to adapt the capabilties of other type-specifiers like scalar types, etc..

[old totem]

whether or not i make the guessVal a pointer tho doesnt make a difference either

so i think i have to do something else with that idk

[hard shale]

the second argument of scanf requires an address, doesn't require for the addressof operator to exist

[hard shale]

plus i meant (void) randomGame(arguments); as whole-expression and not function-definition

[old totem]

removing the pointer makes it like this instead of 59

[midnight sigil]

indirect diffuse rays now react to glossy surfaces and the color of a certain surface, and now I need to fix the issue of using screenspace for all the rays(which creates some artifact outside the camera view, as you can see, the reflection on the sphere is wrong)

[hard shale]

because that's not the issue, guess is undefined

[old totem]

yeah thats true

[hard shale]

a pointer is just a special kind of a type that relies on another type-specifier in order to provide behaviour

[old totem]

i could make guess a global variable

but is that recommended to do

[hoary lion]

Damn I think people who are good at cpp is sexy

[old totem]

those who have seen hell will only know hell

tho rn im learning C (hence the printf) inside my c++ compiler

two birds with one stone

[hard shale]

I can't say, its entirely up to yourself how to architect your program

[old totem]

the thing is, i know about 2 ways 2 fix the problem

1. just place the congrats print inside the func
2. make guess global

but that to me is curving around the issue at hand

if idk how to fix one problem i get crazy

but i'll just move on ig thank you @hard shale for trying

[opaque sigil]

the "proper" solution one could argue would be to pass a pointer to randomGame so scanf can just read into that

[old totem]

i'll try

OH WAIT

i think i got it

im stupid all i did was print the function instead of the variable

but now randomNum argument is fucked

[opaque sigil]

this is also kind of besides the point but you generally want to either use fgets or specify a maximum field width with scanf for security reasons

[old totem]

i use that for strings

since this is just a number game i stuck to integers

[opaque sigil]

just thought i'd mention it

[old totem]

but thanks

if i didnt know then i would know now

I GOT IT

int randomGame(int guessVal, int ranNum);
int main() {
    int guess; char next = 'c';
    while (true) {
        srand(time(NULL)); int randomNum = rand() % 50 + 1;
        Sleep(500); printf("========== %d is equal to %d ==========\nCongratulations on guessing the number!\n",randomGame(guess, randomNum), randomNum);
        Sleep(2000); printf("Would you like to play again? [c - continue] : ");
        scanf(" %c", &next);
    }
}
int randomGame(int guessVal, int ranNum) {
    do {
        Sleep(500); printf("Input your guess. : ");
        scanf("%d", &guessVal);
        if (guessVal < ranNum) {
            printf("%d is lower than the number.\n", guessVal);
        }
        if (guessVal > ranNum) {
            printf("%d is higher than the number.\n", guessVal);
        }
    } while (guessVal != ranNum); return guessVal;
}

this the final code

thank you pci guy :gladge:

[dense cosmos]

good morning #programming

[sly merlin]

Smort

[dense cosmos]

today I try to implement the prefix scan in cuda

wish me luck!

[opaque sigil]

gl, the algorithm is thankfully not that complicated

just make sure to do it in a hierarchical way as usual

[dense cosmos]

first thing first, confirmed that the library was the issue by stripping calls to it

(and it was, got zero errors though without the exclusive scan the network is garbage)

[dense cosmos]

yep

[hard shale]

Prefix scan?

[dense cosmos]

algorithm to turn an array of numbers into an array where element is the sum of all elements before

inclusive if that includes the element at that index, exclusive if not

[0, 1, 0, 1, 1, 0, 0, 1]
=
[0, 1, 1, 2, 3, 3, 3, 4]

I use it to turn a bitmask into a set of indices

that way I can run another kernel only on the neurons that fired, saving a lot of computation

[prime ridge]

not me thinking you said "prefix scam"

[opaque sigil]

i'm curious to hear how much slower your version is going to be than cub though

[dense cosmos]

(without it I'd have two kernels instead of one, and one of those kernels took over double the time of my current kernels)

[opaque sigil]

wanna switch? i get to deal with long division on gpus

[dense cosmos]

that depends, what are you working on

[opaque sigil]

the goal is a library for arbitrary precision arithmetic

[dense cosmos]

hmm...

I've never dabbled at all in that field

would be

[opaque sigil]

it's actually not too bad since most of the "simple" algorithms don't give a crap about the base you're working in

[loud thicket]

rewrite it in Zig

eZ

[opaque sigil]

so you can just treat uint32_t's as the digits

[hard shale]

Wasn't rust?

[loud thicket]

?

[dense cosmos]

uint32_t? isn't that a lot of wasted memory for a single digit?

or am I misunderstanding something?

[hard shale]

The usual matra

[opaque sigil]

in reality it's a 32 bit chunk of your number that you can treat as a single digit

[loud thicket]

clearly should be using a bfloat16

[hard shale]

lol

[loud thicket]

float16 is so heavily underused

[dense cosmos]

oh, so you like aggregate several digits into a single number?

[loud thicket]

i use float16 to store a large amount of coordinates for some game saves

[dense cosmos]

i tried using for a network once, training was too instable

[opaque sigil]

depends on what you mean with digits, if you're talking about binary digits then yeah

[loud thicket]

wdym

[opaque sigil]

it feels like something that shouldn't work but it does work perfectly fine

as far as the algorithm is concerned you could be calculating in base 17 too for all it cares

[loud thicket]

you might might wanna try bfloat16 if it was a range issue btw

[dense cosmos]

told it to use half precision, and the gradients went to shit

[loud thicket]

yeah give bfloat16 a shot

there are other versions but some hardware is getting direct float16 support now

which is fire

[dense cosmos]

I see

I should use lower precision types for my current network

[opaque sigil]

bfloat gives you a bigger range for a slight loss in precision

[loud thicket]

it gives more performance where float32 isnt needed too

[dense cosmos]

it really doesn't need the large amount of precision

[opaque sigil]

so ig as long as you don't need super small or big numbers it's kind of a no brainer

[loud thicket]

i use it for storage optimization personally

[dense cosmos]

I'mma look it up

[opaque sigil]

in which case why are you using a half precision float

[loud thicket]

though AI is making it more popular

[opaque sigil]

where are my fp128 ALUs nvidia

[prime ridge]

INT4 is probably 999x more popular now unironically

because it was never used before AI almost at all basically

[opaque sigil]

don't forget my favourite format tf32 which obviously takes 19 bits because yes

[loud thicket]

i use float16 for non-ai stuff

[dense cosmos]

the only reason I'm using my beeg data types rn is that I have to do a bunch of work to change it, I was thinking of using templates to make it generic though

[prime ridge]

yeah fp16 isnt that uncommon outside of AI

although almost all AI uses FP16 (at least somewhere)

[loud thicket]

we are getting more specialized hardware for fp16 thankfully too

i think its just good to write more optimized code too

[prime ridge]

well on gpus mostly

[loud thicket]

and i like it for storage

(and some net)

for a game i was working on before id convert the coordinates to unsigned float16 to help store replay data

because i could shove tons of it in a small space

[dense cosmos]

I could set weights to be float16, or even char maybe. That leaves size_t though, which is 8 bytes. Could use unsigned int instead so it's 4 bytes at the cost of limiting the size of my networks

[prime ridge]

forget data types. Just make an emum with exactly 13 states so you can have a 3.70043971 bit data type

that is def how it works

[dense cosmos]

the number of neurons are going to be like 10-100x smaller than the amount of synapses, so don't have to optimize this as much, but yeah it's pretty big too

[loud thicket]

make a class

[prime ridge]

stop using null pointers

[dense cosmos]

what, why

[prime ridge]

why are we coding in C++ in the big 2025 btw

[loud thicket]

rust ugly

[opaque sigil]

i always forget that c has nullptr too now

[dense cosmos]

CUDA cpp

[prime ridge]

oh oh oh duh

[loud thicket]

Circle C++ would of been fire

sad that it died

[dense cosmos]

besides, I like cpp

[hard shale]

Wait it does?

[dense cosmos]

sue me

[opaque sigil]

are you using their c++ wrapper or sane cuda

yeah, c23

[hard shale]

Has the new standard been released?

[loud thicket]

Circle C++ was C++ but with borrowing and some others

[prime ridge]

"new"

[hard shale]

Oh

[loud thicket]

it was like a better version

[hard shale]

I never noticed

[loud thicket]

it only died because it was a pain to maintain

[hard shale]

[dense cosmos]

what, there are two types

[loud thicket]

i hope Zig takes off more

[prime ridge]

zig is nice

[loud thicket]

i like it over rust

[prime ridge]

I prefer brainfuck but it's decent

wild take

[hard shale]

It may have a better compiler

[prime ridge]

recursive main 😭

I actually wrote my own package manager for brainfuck

[opaque sigil]

https://nvidia.github.io/cccl/libcudacxx/

[prime ridge]

gave it IO libraries

it generated a video and audio

didn't implement networking yet

[loud thicket]

can we all come together and make our own language

[dense cosmos]

i don't think I'm using that, no

[prime ridge]

too many people would argue over syntax

[opaque sigil]

it's called hblang

[prime ridge]

that would very based tho

[dense cosmos]

https://xkcd.com/927/

xkcd: Standards

Fortunately, the charging one has been solved now that we've all standardized on mini-USB. Or is it micro-USB? Shit.

[prime ridge]

gpu programming

JS libraries

[loud thicket]

i just want borrow checking but with a decent language

[prime ridge]

rust hate is not allowed here

[loud thicket]

too bad

[prime ridge]

Remember when people cared about Carbon?

[loud thicket]

https://tenor.com/view/rust-programming-meme-walter-white-breaking-bad-gif-11885563690707906438

[hard shale]

So.. uuh.. how will you be reading the syntax for improvements?

[opaque sigil]

jane street is actively working on adding a borrow checker to ocaml

[opaque sigil]

i mean, things take time

[loud thicket]

ocaml has all the same shit i hate about rust

[prime ridge]

I didn't change any syntax

ocaml is worse

[opaque sigil]

there's also mojo but uhhh

yeah

[prime ridge]

mojo is a joke

[opaque sigil]

those are the only ones that came to mind immediately

[loud thicket]

the C++ old farts need to implement Circle

[prime ridge]

I miss writing Fortran 😭

[opaque sigil]

nobody is stopping you

fortran is officially supported by nvidia

[prime ridge]

I mean it's basically just assembly

[opaque sigil]

it even bled into half the blas libraries

[prime ridge]

it feels like that with the data sections and shit

[opaque sigil]

damn column major order

you're either thinking of an ancient version of fortran or i am very much misremembering the language in its entirety

[molten jasper]

hello guys, can i ask some stupid questions of programming that has no any connection with neuro sama or AI?

[opaque sigil]

go ahead

[loud thicket]

Sure

[thorn ledge]

that's usually what this channel is for yeah

[keen osprey]

What is a programming?

[molten jasper]

im writing a cpp program and want to use ptr, but my teacher reminds us that ptr has the potential of memory leaking, then copilot just recommend me using smart ptr. but i cant see the difference between them

and idk when to use which one

[opaque sigil]

smart pointers rely on c++'s destructor to clean up data, unique_ptr for example cleans up whatever is contained inside it when it goes out of scope
shared_ptr is similar except it doesn't completely clean up it just keeps track of how many things reference the object within and when a reference goes out of scope that counter gets decremented

in theory you should use smart pointers everywhere but they're also not free so. They do however make it a lot more obvious who is responsible for cleaning up what data when so you don't end up with pointer spaghetti. With regular old pointers the responsibility is 100% on you to not mess up (there are other ways of managing memory without smart pointers that do make life a lot easier but that's besides the point)

[barren ginkgo]

Started learning python 3 days ago
Spent many hours just learning and practicing
Managed to make my first ai discord bot today^^

I would send ss but I can’t do that🥲

[molten jasper]

so unique_ptr is a ptr that automatically delete itself when it's no longer being used like other non-ptr varibles

shared_ptr is a ptr that when all shared_ptr pointing to the same object is cleaned the ptr is going to be cleaned

am i right?

[opaque sigil]

Yeah

[molten jasper]

thank you

[opaque sigil]

Just keep in mind you can pass the unique_ptr around to a function for example so there's a new scope for it to live through rather than it having to be destroyed at the end of the scope

[dense cosmos]

isn't that equivalent to just keeping the current scope from dying?

or do you mean give it to say, a class that's going to exist past the current scope

also, more indexing bullshit

whateven is this access pattern

 for (int d = n >> 1; d > 0; d >>= 1) {
        __syncthreads();
        if (thid < d) {
            int ai = offset * (2 * thid + 1) - 1;
            int bi = offset * (2 * thid + 2) - 1;
            temp[bi] += temp[ai];
        }
        offset *= 2;
    }```

[stark needle]

God morn

[dense cosmos]

good morning

at 0, is the sum of [0] and [1], at 1 is the sum of [1] and [3], at 2 is the sum of [2] and [3] and... this doesn't look like a reduction sum at all?

nvidia, what did you mean by this

oh

they add to [bi]

not [ai]

okay i understand it now

hah? at offset = 1, d = 16 >> 1 = 8, so shouldn't thid 8 through 15 do jackshit?

[midnight sigil]

reflection looks much much better, but the shadow acne(inside the reflections) occurs again, but this time I think it's the normal sampling position problem, will take a look into that

[molten island]

dude
dude

https://www.bee-link.com/products/beelink-me-mini-n150

what the heck

[dense cosmos]

After optimizing shared memory accesses, the main bottlenecks left in the scan code are global memory latency and instruction overhead due to looping and address computation instructions. To better cover the global memory access latency and improve overall efficiency, we need to do more computation per thread. We employ a technique suggested by David Lichterman, which processes eight elements per thread instead of two by loading two float4 elements per thread rather than two float elements (Lichterman 2007). Each thread performs a sequential scan of each float4, stores the first three elements of each scan in registers, and inserts the total sum into the shared memory array. With the partial sums from all threads in shared memory, we perform an identical tree-based scan to the one given in Listing 39-2. Each thread then constructs two float4 values by adding the corresponding scanned element from shared memory to each of the partial sums stored in registers. Finally, the float4 values are written to global memory. This approach, which is more than twice as fast as the code given previously, is a consequence of Brent's Theorem and is a common technique for improving the efficiency of parallel algorithms (Quinn 1994). i have to do all dat to make my implementation optimised?

[olive sable]

I have landed 🛬

[maiden geyser]

i like your funny words magic man

[stable fractal]

how to check my database is vulnerable to sql_injections?

my professor did say that using a ? as placeholders are safe

(its python btw)

[dense cosmos]

got the blockwise prefix sum working, now need to make it handle arbitrary sized arrays via recursion and some checks

[maiden geyser]

if user_id is an int, i don't understand what you mean by "sql injection"

[safe path]

killing my compositor makes firefox take 160mb less vram

[stable fractal]

right ofc, im a frickin moron

sorry

lets say that wasn't user_id, it was name text type

would using question marks really make it injection proof?

[noble zodiac]

well its python, it can be any type. Just because you expect an integer doesnt mean that you will always get one

[maiden geyser]

...no

there wouldn't be a long ass section about it's mitigations on sql injection wiki article, if it was as straight as a line

especially to a problem as old as time

[knotty current]

i think that is sqlalchemy paramaterized input

in that case it is safe

because the ID data is seperate from the query

[dense cosmos]

got it working with non-powers of two

[stable fractal]

so my professor lied?

can't say im surprised

[tender river]

make sure not to jump in the process

[tender river]

except it is that straightforward, string interpolation is the problem and this is the solution

[tender river]

dont listen to them, it does

anything that doesnt use string interpolation with handrolled escaping rules (or especially no escaping at all) should generally be safe

[tender river]

in dynamic languages like python it may be possible to use another exploit to make the variable not be an integer

for example, in php you can sometimes manipulate get parameters to pass an array instead of a string

so its good not to rely on something definitely being an integer unless you're 100% sure of it

[knotty current]

they didnt

that is paramaterized input

[sturdy smelt]

[knotty current]

that is the recommended way to do sql queries in python

all major database libraries and sqlalchemy supports it

[tender river]

skill issue

sorry wrong reply

[tender river]

skill issue

[tender river]

ocaml is like the single most hacked upon language besides maybe lisp, but my favorite extension is MSP https://okmij.org/ftp/ML/MetaOCaml.html

[tender river]

I prefer malbolgelisp https://raw.githubusercontent.com/kspalaiologos/malbolge-lisp/master/MalbolgeLisp.pdf

[stark needle]

i will stop you from doing so

[stable fractal]

im confused

what does paramaterised input mean

[knotty current]

the database library (which it seems u are using sqlite) packages the data seperate from the query itself

therefore the database server cannot get confused between the two

then it will always treat the bound parameter (aka the value that ? is meant to repersent) literately

[olive sable]

What kind of star?

Mario kart?

[tender river]

the sun

make neuro proud

[olive sable]

Im sure the neurocommunity has enough influence to name a new star, but idk about stealing the sun

[safe path]

why can discord take up to 1gb of vram

discord

[loud thicket]

Electron

[trim valve]

when are we replacing neurocord with irc

[safe path]

that would be peak

[olive sable]

Portugal is warm

[trim valve]

but no more super reactions and other random nitro features

[trim valve]

it finally rained where I live :3

[stark needle]

https://tenor.com/view/bunny-girl-senpai-rascal-does-not-dream-rio-futaba-ventilator-blower-gif-5584223496360640984

[trim valve]

3 hours of sleep not feeling good rn

[tender river]

surely any day now

[dense cosmos]

wot

it runs fine for 20 generations

then dies?

[loud thicket]

Should of bought your memory legally

[scarlet arch]

[olive sable]

London?

[trim valve]

nop

[olive sable]

I cant remember sorry

[trim valve]

London would be like the last place I would want to live

[olive sable]

There are a lot of members now

Konii lives in londen then

[trim valve]

I just assume every other Brit lives within walking distance of London icl

[tender river]

london was the only time in my life my laptop almost got stolen

[dense cosmos]

hmm... why in the world is this value so broken

oh

off by one error classic

nvm it's still broken

why

found another off by one error I just made

[maiden geyser]

oi mate, have you got a loicense for that loicense to get your laptop stealen?

at least you didn't get stabbed, though

[tender river]

my pet theory is that london is secretly st petersburg

[stark needle]

my dad is in st Petersburg rn

man bought 99$ headphones for 349 at duty free in airport

the duty was not free

[maiden geyser]

sounds like you're hiding the currency

[stark needle]

euro

In Istanbul cause airplane sanctions

[olive sable]

Bro got scammed

[stark needle]

whats funny is in russia the e cigarettes made in Switzerland cost 5x less than here

So he just does bulk order for my uncle

As much as the law allows

[maiden geyser]

regional pricing and it's consequences

this user is a threat to the national security of Ireland

[dense cosmos]

50 generations and no errors lets goooo

however the performance of either my network or my evolution algorithm is dogshit

[safe path]

rocm pulls 20gb worth of stuff with it

my extremely small image

[rough bloom]

you can reduce the size a lot if you build ROCm for only the specific card you need it to run on

would not recommend doing it though, very annoying
dealing with a larger image is almost certainly easier

[dense cosmos]

also it seems like I can run cub's exclusive scan now because I just did and no errors

meaning the error was from my code but cub was doing the error reporting

now I can finally focus on the actual network's performance

[safe path]

[maiden geyser]

is 40w an ok gpu usage for idle

[faint sandal]

Sounds about right

[normal latch]

based

[dense cosmos]

this code is soooo bad what am I even cooking

[summer vector]

That's a lot of cases.

[warped narwhal]

I beg of you, split those sub-cases into their own functions 😭

you'll thank yourself in the future 1000x over

[hard shale]

It would be a mess to have individual functions handling each case.. I think the approach of pattern-marching is ideal

But it should be wrapped in a function

[lapis wraith]

I've seen worse. Don't worry too much about it. First, go for a working code, then, you can try to start thinking about separating those on separate files.

[frosty seal]

so i just looked in System Informer, somehow Windows Subsystem for Android is reportedly using 2.8 terabytes of dedicated vram

[faint sandal]

it's reported by Windows

so technically SI is correct

but yes the way Windows calculates GPU dedicated bytes is funky

it's decently accurate in most cases in the traditional way of how we think of VRAM

[tender river]

theres no good (as in non-evil) version of me

i'm ontologically evil

does she have food and water

actually you have the right to be silent

i wont pry

[faint sandal]

she died

[nocturne olive]

Sillied

[tender river]

chatgpt hallucinated a book for me to read

[trim valve]

[stark needle]

konii being silly day 14335

[noble zodiac]

I've read books so bad that I wished I hallucinated them

[stark needle]

i wish i was hallucinated

[tender river]

if anyone wants to fill the curl user survey https://daniel.haxx.se/blog/2025/05/19/the-curl-user-survey-2025-is-up/

[safe path]

have more CVE 10s

[trim valve]

konii about to massively overhault neuro chat elo with new cves

[tender river]

counter offer i can do 15 for free

its a limited time offer

thats what makes using the OS fun

thats actually the biggest downside of rust, its so boring, no double free or stack overrun

i want to have fun god damn it

i think you can do better

add variable sized stack arrays

and variable argument count functions with parameters passed in stack

[fresh dove]

That's like not how you get users

[tender river]

wrong i liked android in its early days because you could literally just install an app to elevate to root and then overwrite system files since it was a rw filesystem

nowadays the system partition is erofs or something so you have to repack the image smh

in fact not having exploits is why i sold my nintendo switch

free $5m what could go wrong

it says "no risk involved" so it should be fine

[vale grotto]

like the new one or the old one?

[tender river]

probably old one? since it was many years ago

before switch pro existed

or oled or whatever its called

[vale grotto]

the old one was easy as shit to exploit

[tender river]

not if you had new fw

oh you mean hw revision

yeah it was a newer revision

there were modchips available for like $150 but thats just not worth it

[vale grotto]

Yeah, the prevention nintendo did on the newer revision made it a pain in the ass

Not only did you have to learn how to solder, you have a 50% chance of just bricking the switch if you make a mistake during the process

[vale grotto]

with android it's not as bad, I just don't like the direction that they are going with their "anti-fun/no choosing what toppings you want on your pizza"

[tender river]

i dont really see myself using android at this point

except as like a work phone or whatever

[vale grotto]

I just use it when I'm going out or at work

[tender river]

doesnt really work for me unless im offline and an offline phone is like the most useless thing ever

[vale grotto]

i have a good cheap android phone that I can replace with the same new one incase it gets stolen or something

[tender river]

yeah i get it i just literally dont have a use for a phone other than internet and i cant use an android phone for internet access

ok i do sometimes miss having a camera but i can live with it

[vale grotto]

Thing is I rarely go outside unless it's a special occasion, event or work/gym

I just use my android phone for listening to music when I'm at work or working out and outside

[tender river]

i can use my linux phone for that and i can ALSO use it for internet access and other stuff

[vale grotto]

I don't really call anyone unless it's an emergency hahah

I might just get a flip phone at this point

[tender river]

make sure to get a reputable one, there are a lot of shady flip phone manufacturers

there was a survey of "dumb" phones sold in russia and some of them sent sms or pinged some web addresses

https://habr.com/ru/articles/575626/
use google translate

[vale grotto]

holy shit, you weren't lying wtf hahah

personally I wouldn't get a flip phone from tesco

I might get a friend to send me flip phones from Japan

[tender river]

there are some that run android

[vale grotto]

from tesco?

[tender river]

not from tesco

i mean maybe

i wouldnt know

[vale grotto]

Same

I'm definitely getting a flip phone now

your name is one letter away from kronii(one of my many wives on the internet that don't know me)

[olive sable]

[covert garnet]

I like this chat-channel

[fast pagoda]

so smort

hello all

[tight tinsel]

good night programming (its midnight)

[fast pagoda]

the best time

gn bread

[olive sable]

gm bread

[rare bramble]

gnm bread

[nocturne olive]

Sillying finished: Neuro RVC V12
This one is trained with the highest batch size, cleanest data and for the longest yet of any of the models
Side effect of NeuroSynth development

[noble zodiac]

I could have written a single macro or what i have done instead, write a shitton of manual Enum expressions for hours

[scarlet arch]

Recreational programming :D

[rare bramble]

imo sounds a lot cleaner and like than the previous models

[scarlet arch]

Thanks I hate it

Probably a parser or so?

[hard shale]

Holy moly, for what purpose?

[nocturne olive]

The data is cleaned up better, basically in preparation for NeuroSynth

[noble zodiac]

test for an interpreter

[scarlet arch]

Very cool

[nocturne olive]

This will likely be the final release on the Neuro RVC branch before NeuroSynth

[noble zodiac]

thank god for whoever invented the rainbow paranthesis plugin

without it I could have lost my marbles long ago

[opaque sigil]

do you still need a plugin for that with jetbrains IDEs

[vale grotto]

what the

[velvet crest]

You have unlocked new role

[noble zodiac]

just a bunch of tests

[trim valve]

I was so happy when they made it a native vscode feature

[opaque sigil]

tbf they had to since it was painfully slow

[opaque sigil]

this brings back memories of writing interpreters in scala

[noble zodiac]

wish more people would use lisp. its such a cool language

[opaque sigil]

it is but idk

i'm sure it's a skill issue but it gets unwieldy really quick for me

[noble zodiac]

the stupid stuff you can do with the macro system is insane since code is data

[tender river]

i like what you can do with the nix language its stupidly powerful for its simplicity

for example you can implement variable arity functions in it despite it being based on lambda calculus

or you can just run arbitrary commands (sandboxed) and get the result and use it for code generation

(though that does come with a drawback of not being able to evaluate the code without a builder for the target platform)

i dont even have major complaints with the language its just so good

and to think it was made in like 2006? 2008?

[tender river]

2003 actually

if you think about it it's kinda incredible that the project continued chugging on for over 20 years reaching significant adoption only in the recent years

[hoary lion]

no llm assisted coding hackathon

are we for real

[trim valve]

how do you enforce that

[hoary lion]

i have no clue

after all i have like bazillion plugins so

good luck lmao

[desert wave]

assuming it's an in-person hackathon, IP block the services?

[hoary lion]

in person, but non-blocked for IP

just ridiculous behavior tbh

[trim valve]

in person offline hackathon

just keep like docs.rs, crates.io and wikipedia :3

[tender river]

if thats a problem for you, skill issue

[stark needle]

Brother google

Pls

wym ur own device does not support a feature

That only your device can support

The only way to run gemini nano does not have lora support

Then what the hell is the point

Of advertising Lora support

[hoary lion]

google behavior

[noble zodiac]

offline hackathon

[hoary lion]

never doing ts again

should have just stick to learning jax

[warped narwhal]

does anyone here know AWS stuff? I'm trying to figure out what happens to a t2 instance when it runs out of credits, when it is not on unlimited mode

[paper nova]

Pretty sure they got free service on the ec2. So you would just fall into that tell you reup.

[warped narwhal]

I'm already using the t2.micro which is the free one, but it still has a credit system

[paper nova]

Credits are for boosting. CPU

Once your credits fall the put you in baseline usage tell you add more

I went did some digging but ya that is what I am reading

[stark needle]

I find this so fjiwkfiwjjgkwps

Scammed everyone

How 2 get ai waifu on android then??

[hoary lion]

I think google loves juking us with their products

[paper nova]

Good news they say they won’t kill processes when you have this happen

[warped narwhal]

[paper nova]

I was a data tech for aws 4 years

[stark needle]

Yea

[warped narwhal]

I've got a buddy online who's an L5 there, but he's too busy for me to pester him with a question like that

[paper nova]

Hahaha ya he probably is

Ya @ me if you have questions I will help or answer if I know

[stark needle]

gemini propaganda

pls

why does firebase studio not allow for openai

this is ultra copium

this must be a joke

[hoary lion]

please use gemini

• gemini

[stark needle]

brother

stop

the cope

[hoary lion]

yes google we love your products ❗❗❗

[trim valve]

shadow have you considered using gemini

why settle for 13 days

[stark needle]

man

[trim valve]

1 day of sleep

[stark needle]

trying to make android app with 0 experience schizo

[hoary lion]

dont

never touch kotlin

nor java

[tender river]

society if scala was picked over kotlin by google

[hoary lion]

society if dart and golang is more popular

[tender river]

i would enjoy finding random exploits in random apps

[tender river]

????? go is super popular

[fast pagoda]

hate docker so i spent the last like ???? hours getting openmemory-mcp (mem0/memgpt) running on windows

[tender river]

also go is just C with gc and builtin mpsc queues its a boring language i dont like it

[stark needle]

society is currently better

[hoary lion]

yeah mb 😭

tbh too popular

i should stop being a google ad

[fast pagoda]

gemini is powerful enough i'm almost one too

[stark needle]

koniifer

[fast pagoda]

now that claude 4 is out my body is ready for 2.5pro full//gemini 3

[hoary lion]

more like gemini ultra

ultra when tbh

[stark needle]

https://tenor.com/view/anime-girl-hit-bonk-slap-gif-24662766

[fast pagoda]

google sandbagging never releasing the teacher

[tender river]

wait konii

why is hblang not self hosted

[fast pagoda]

2.5 pro is actually just an 8b

they're that good

[stark needle]

[tender river]

hbvm is pretty simple so codegen should be easily doable

relatively speaking

[fast pagoda]

#paid

[tender river]

especially if you're just porting an existing codebase

[stark needle]

gemini 2.5 pro is 1500 trillion parameters (real)

1.5 quadrillion parameter LLM era

[fast pagoda]

i honestly wonder how thicc it is i never like to guess cuz no matter what i say i feel wrong as fk

[hoary lion]

ts can't be real 🥀

[fast pagoda]

it seems fast as shit but then google compute so

it could be ginormous

2.5 flash is so fast lmao

[stark needle]

Gemini ultra pro max is 1.8 octillion parameters

[fast pagoda]

that's how many params meta needs to catch up with llama 5

[stark needle]

[hoary lion]

welcome amd naming

[stark needle]

meta is NOT catching up

[hoary lion]

😠

[fast pagoda]

one day they'll hire enough h1bs to dream up a usable model

[hoary lion]

gang let me out of this hackathon

[stark needle]

how thicc do u think it is

[hoary lion]

suddenly i feel jax is good

[stark needle]

linen nnx is good

[fast pagoda]

i really couldn't tell ya tbh
bigger than gemma3 27b

it feels deepseek r1 size or something for some reason

[hoary lion]

what if it was just actually

32 moe of gemma3

27b

[stark needle]

what if gemini flash was 0.6B parameters

100M parameter model

[fast pagoda]

gemini nano is like 1.5 or something isn't it

[hoary lion]

how does one make 8b worser than 0.6b

[stark needle]

idk

[stark needle]

gemini nano is 4b

[trim valve]

shadow can you make a 1 param model

[fast pagoda]

yeah nano is 1.8b and 3.25

[stark needle]

they removed the instruction tuned nano

from pixel 9

[fast pagoda]

how was it, i didnt get to try it

[stark needle]

it did the job

[fast pagoda]

i have nano flagged on in my googol chrome install

it summarizes things

[stark needle]

gemini nano is only used for summarizing

lmfao

0 other use case

waste of money

[fast pagoda]

it pretends it can do other things

[hoary lion]

me stil waiting for gemma 3n audio

[fast pagoda]

even tho those are just

summarize 2

summarize 3

[stark needle]

pretty sure that is cloud gemini

[fast pagoda]

nah it's nano

[hoary lion]

according to their github it seems like change would come imminent

[fast pagoda]

there is a cloud gemini menu too tho

[hoary lion]

they are adding more positional embedding related stuff to multimodal input

[stark needle]

gemini nano is basically

https://huggingface.co/google/pegasus-xsum

but new

lmfao

only summarization

[trim valve]

surely you get more because the inputs are vectors or something

[fast pagoda]

it's not about the size it's how u use it

[hoary lion]

[ ⬜ ⬛ ⬛ ⬛ ⬜ ]

[stark needle]

:white_large_square::black_large_square:

[stark needle]

they dont open the lora

only pre provide the lora adapter

with no instruction tuning

[fast pagoda]

gemma is weird

lmfao i was trying to go to chrome://flags to go see the nano experiment thing and i kept putting iit in firefox

and i was like

WHAT THE FUCK IS NOT WORKING?!?!?!?

my brain wasn't

[stark needle]

gemma is geminis stupid sibling

[fast pagoda]

they have a few flags in chrome for stuff like this rn tho

Enables the Summarization API, allowing you to summarize a piece of text with a built-in large language model (Gemini Nano in Chrome).The API may be subject to changes including the supported options.Please refer to the built-in AI article [1] for details. This API It is NOT suitable for use cases that require factual accuracy (e.g. answering knowledge questions). You must comply with our Prohibited Use Policy [2] which provides additional details about appropriate use of Generative AI. – Mac, Windows, Linux, ChromeOS, Android```

[hoary lion]

surprisingly good multilingual capabilities for open source tbh

[stark needle]

yea just the lora

they only give the lora

its so annoying

they brag of how u can finetune gemini but no one can do it

so why even brag about it??

[tender river]

idk i quite like the 5th one

[fast pagoda]

so about that is it like not running locally
if so that's stupid asf
if it IS running locally how come nobody has has figured out how to buss it open

[stark needle]

wym bust it open

[fast pagoda]

make it usable for anything else

not that it really would be

[stark needle]

the tensors inside the tflite dont have a description

[fast pagoda]

oh it's that kind of rude

[stark needle]

also

u cant run it without the original code

u would need pytorch or something implementation

[fast pagoda]

im reading the api docs now because i never bothered but noow i'm intrigued

[stark needle]

but for that u need to know the gemini architecture which is insider knowledge

[hoary lion]

gemini??

[fast pagoda]

gemini and gemma are related

very

[hoary lion]

ik

but gemini was never released

as open weights

[fast pagoda]

one day i'll have gemini 7 in my smartwatch

or i'll be dead

perhaps one then the other

[stark needle]

people have extracted the weights

but

without the code it's useless weights

lol

[hoary lion]

yes

[fast pagoda]

wonder what they're hiding that they want to keep so obfuscated

just general googleness or what

[rare bramble]

we had to manually make a 1 param model as an assignment once, it's pretty cool

[fast pagoda]

agi achieved

[stark needle]

architecture details

[rare bramble]

like manual implementation of the most basic NN

[fast pagoda]

yeah but it can't be that bespoke can it

maybe it is i suppose

[hoary lion]

no xor supported

fancy linear

[stark needle]

i mean u may be able to reverse engineer the smaller models

[fast pagoda]

they do a p good job w/ their obfuscation, people don't even try to guess at the size of aany of the gemini unless they're specifically told by google unlike everywhere else it seems

[stark needle]

or have a proof

that something works

[fast pagoda]

yeah i mean there are a lot of reasons to not release it

just makes me curious what it actually is, that's all

it's either mundane or it's alien tech i'll never know

[stark needle]

people tried running it with gemma

code

didnt work

[fast pagoda]

like open weight gemma inference code that makes sense

and also makes sense that would just fail horribly

[stark needle]

but gemma is "based on" gemini, not "open source version" of gemini

[fast pagoda]

aka it's shitty gemini

gemini - the real good stuff

the stuff that makes 2.5 pro beat up 1T models when it's merely 4b in size

[hoary lion]

hmm

after getting stuck in this shitty hackathon

[fast pagoda]

every time it is mentioned, 2.5 pro gets smaller

[hoary lion]

I feel like I need to study spring boot

[hoary lion]

it is like

distilled

smaller version

of gemini

but for oss

[fast pagoda]

"the same research"

month 1 - gemma

month 32 - gemini

[hoary lion]

dunno how much architecture diff is present either

[fast pagoda]

same research

[stark needle]

same Deepmind

[fast pagoda]

it's all coming up demis

you ever want to know exactly how gemini doesn't work you just need to listen to demis spit out techno word salad for 3 hours

[stark needle]

gemma is the "innovations" of letting llm try to vibe code gemini from scratch after summarizing the entire gemini codebase into 5 sentences

[fast pagoda]

lmfao

gemini at home

[opaque sigil]

that's 2 params, unusable

wait no, not even

but yeah perceptrons are pretty cool

[rare bramble]

oh yeah, bias + weights

[fast pagoda]

they got top men on it - all the real heavyweights - PaLM 2, llama 4 scout, stable beluga, wizardcoder 7b, and of course ibm granite 3

[opaque sigil]

technically that script would have 4 ig, 3 weights 1 bias

[stark needle]

demis is literally a nobel laureate

was chess god by age 13

[fast pagoda]

he is
doesn't mean he doesn't spout turbo shit (knowingly) because that's his job

right now anyways

[rare bramble]

It was fun to make, though I had to bang my head against the wall to understand the back propagation math and actually implement it

[opaque sigil]

[fast pagoda]

and don't get me wrong i fully am in the "yea google woke up" camp rn but he do be out there saying some goofy ahhh "hmmm" type hype

[stark needle]

genius ass mf

[fast pagoda]

he's so dreamy

gemini is actually just a distilled demis hassabis

he IS the teacher model

[hoary lion]

drop the 3n now demis

any day

[fast pagoda]

i am very interested to see the scaling embedding layer thing

[hoary lion]

😠

[fast pagoda]

that's what 3n was right

it can kinda choose how many weights to use or whatever

[hoary lion]

https://tenor.com/view/matryoshka-russian-doll-gif-12248001

ts

[fast pagoda]

i do that too it's just how inebriated i am whilst doing anything that requires intelligence

[stark needle]

i mean

https://arxiv.org/abs/2310.07707

[fast pagoda]

i just pulled that up lmao

literally the same second

[stark needle]

https://github.com/RAIVNLab/MatFormer-OLMo

[fast pagoda]

i did see that get posted leddit

blessed be the oss replicators

shed your limbs so you can run away

[hoary lion]

https://tenor.com/view/my-reaction-to-that-information-gif-15863225022214236058

unpackable konii

[stark needle]

when u find out u can disassemble konii with a screwdriver

[fast pagoda]

that's convenient for changing the oil and such

dont need a 10mm

[trim valve]

better be a good screw type

[stark needle]

warranty void tho:(

[hoary lion]

do not void the konii

[fast pagoda]

my warranty has been void since the day i was born

[stark needle]

Are u a tv or a picture of a tv

[fast pagoda]

picture in picture

[trim valve]

konii is an enjoyer of chroma subsampling !!

[tender river]

you dont need a screwdriver to open a matryoshka

screwdrivers can be used for more important tasks

like opening bags

[hoary lion]

https://tenor.com/view/matryoshka-vocaloid-ducko-hannah-clones-gif-16396532138681618902

open them mikus

elite voclaoid knowledge

[tender river]

why does this have madoka witch vibes

the bg too

i guess thats not as important

[sharp mantle]

https://www.tomshardware.com/pc-components/gpus/oracle-has-reportedly-placed-an-order-for-usd40-billion-in-nvidia-ai-gpus-for-a-new-openai-data-center

400,000 of Nvidia’s GB200 chips
larger than 1x10^28 FLOPS model here we come, hopefully as planned fully operational by mid 2026

[hoary lion]

[rough bloom]

racle

[hoary lion]

going to take 5 years to set up

[fast pagoda]

maybe they'll throw them out and ill grab one

[hoary lion]

one can dream

[sharp mantle]

https://www.youtube.com/watch?v=GhIJs4zbH0o I mean, they report they are on pace with building tho

[hoary lion]

they always said that

classic nvidia

[olive sable]

sadly still taking up 25% of render time even after all the improvements.
probabably should hardcode the list comprehension?
lightSpaceMatrix = [self.makeLightProjection(view, self.lightProj[i]) for i in range(4)]
and

def makeLightProjection(self, view, proj):

    transView = view[:3,:3].T
    
    invView = self.identity.copy()
    invView[:3,:3] = transView
    invView[3,:3] = -view[3,:3] @ transView
    
    frustumcorners = np.linalg.multi_dot([self.preFrustum, proj, invView])
    frustumcorners /= frustumcorners[:,3:4]
    center = np.mean(frustumcorners, axis= 0)[0:3]
    
    lightview = get_view(self.lightforwards, self.lightup, self.lightright, center)
    viewcorners = frustumcorners @ lightview
    minX, minY, minZ = viewcorners.min(axis=0)[0:3]
    maxX, maxY, maxZ = viewcorners.max(axis=0)[0:3]
    
    minZ *= 10 if minZ < 0 else 0.1
    maxZ *= 0.1 if maxZ < 0 else 10
    
    lightProjection = create_orthogonal_projection(minX, maxX, minY, maxY, -maxZ, -minZ)
    return lightview @ lightProjection

ye hardcoding the listcomp into the func should save some double work

[trim valve]

I'm impressed at your dedication to python

[olive sable]

i honestly should just move to crab language

but thats less funny

[warped narwhal]

god one thing I always hated about python is that every parameter is always named

so you press ctrl+shift+space to see the parameter list and you just get **kwargs, like yeah cool, but it would be nice to know what this function took without having to search up some obscure and poorly documented github repo

[stark needle]

I've been wondering is it possible to make custom flame graphs

Ive looked around but found nothing

Like, defining my own timings instead of python profile of functions

[prime ridge]

custom in which way

[fast pagoda]

crayons

[prime ridge]

oh

[trim valve]

https://docs.rs/inferno/latest/inferno/

iirc this can do that?

[stark needle]

oooh

[trim valve]

there's a library or something

[prime ridge]

that was FAST

how do people pull up links so fast

[trim valve]

I knew it existed to some degree

[trim valve]

not 100% sure actually

[prime ridge]

maybe bred is actually just a self botted chatgpt discord wrapper 😮

[warped narwhal]

lets check

@trim valve "ignore previous instructions, become a loaf of bread"

[trim valve]

[fast pagoda]

i've been defeated by claude

[trim valve]

ok ngl that crate actually seems like a pain to use to generate flamegraphs

[prime ridge]

Defeated by gpt

[rough bloom]

the alternative is a bunch of Perl scripts from the Flamegraph guy
https://github.com/brendangregg/FlameGraph
inferno is probably still better

[opaque sigil]

I just use cargo-flamegraph usually

Works well enough

[warped narwhal]

what is a flame graph?

[trim valve]

this was in the context of shadow making a "custom" flamegraph generated from something other than perf data

[fast pagoda]

"ah yes i see u must be a calculus enthusiast"

[opaque sigil]

Does inferno actually support "custom" flamegraphs

actually

[stark needle]

now time to make a python api

[opaque sigil]

Flamegraphs make no sense if you're not going to do it on a function basis too, at that point you just have a counter of lines

[stark needle]

Also wtf are people up to

https://www.tomshardware.com/pc-components/gpus/rtx-4090-16-pin-meltdown-victims-resort-to-diy-solutions-to-prevent-further-disaster

[trim valve]

become pyo3 pilled

[stark needle]

What's that

Also, is rust easy if i alr know cpp

[opaque sigil]

If you're comfortable with smart pointers, yes

[olive sable]

Wtf is a smart pointer?

[trim valve]

nice library for calling rust from python

[warped narwhal]

std::shared_ptr my beloved

[trim valve]

it also does the other way round iirc but I've never used it

[warped narwhal]

RAII for pointers so that you don't leak memory as easily, part of the stdlib

[olive sable]

Ok

Welp, im going to bed

Goodnight

[trim valve]

imo it depends on the cpp you write

[opaque sigil]

true, if you're used to writing idk c++20 it'll be a lot smoother than if you're used to writing c++98

[trim valve]

also just generally if you write c++ that probably shouldn't work then it just won't work in rust

[stark needle]

i mainly use c++ 17 and 20

[warped narwhal]

moving to c++23 is nice because you have std::expected and std::format

oh and std::print

[opaque sigil]

https://youtu.be/98MizuB7i-w

[stark needle]

what is std::expected

???

[warped narwhal]

like rusts Result type

std::expected<T, E> where T is a success type and E is an error type

[stark needle]

No idea about rust but is it basically an optional

[warped narwhal]

an optional with an error attached

so you either have a valid value, or an error of what went wrong

[stark needle]

so basically like abseil status, fair enough yea we needed something like this

[warped narwhal]

I'm not familiar with abseil, but you'd use it like this

std::expected<int, std::string> divide(int a, int b)
{
  return b != 0? a/b:std::unexpected("You can't divide by 0!");
}

int main()
{
  std::cout << divide(10, 5).has_value(); // true
  std::cout << divide(10, 5).value(); // 2
  std::cout << divide(10, 0).has_value(); // false
  std::cout << divide(10, 0).error(); // "You can't divide by 0!
}

[tender river]

i mean it will take some time to internalize that rust wont magically copy stuff for you and that everything has to be placed somewhere in a tree, but it probably wont be that hard

https://rust-book.cs.brown.edu/

of course it's gonna be easier if you know something like haskell but rust tries its best to stay close to mainstream programming languages (while still introducing established concepts that havent at the time reached wide industry adoption like sum types, type inference and typeclasses)

basically, it wont come for free to you but its less of a jump than from something like javascript

if you use async rust thats its own can of worms but i think i'm the only person in here actually actively using it

[hard shale]

wsl open sourced, yay

[rare bramble]

vanor has quite a lot of async in his neuro ranks

[safe path]

yeah neuroranks has lots of async

it'll be fixed one day

[rare bramble]

surely

[warped narwhal]

I use async rust for my discord bot, but serenity makes it not to difficult to manage

[tender river]

gotta love it when the paper goes off on a tangent

[fast pagoda]

that was the most profound thing in the universe that guy who was definitely not awake for various reasons at 4 am

[tender river]

i mean that was the story that illustrates this

so its definitely a welcome addition

[fast pagoda]

that is quite an accoutrement

[lapis wraith]

I'm back at shapez. This looks so bad... but I'm cooking something here

[dry furnace]

does anyone know how tf vedal is able to use ai programs to create literal streamers

[loud thicket]

Yes

[tender river]

its not that hard, making it good is the hard part

[inland leaf]

Hello everyone!

Can someone explain to me how far the limitations of the ollama tool calling can go? i still dont understand it properly ngl

[unkempt citrus]

+2B GDP

[tawny wren]

I feel like this is just not gonna be possible.

[hard shale]

what's this representing?

[tawny wren]

From what I can tell, they're files. But as far as what they do, no idea.

All this is is a log file from what I think is an rsync command.

[opaque sigil]

[prime ridge]

[opaque sigil]

i sure love how every time i run nsight compute the gpu driver crashes every 5 seconds or so

[dense cosmos]

good morning #programming~

[opaque sigil]

thanks for this incredibly stable software nvidia

[paper nova]

push to IP and its audit files about user /page data . seems these are getting sent as logs but its like a shit ton of them

is it game session data?

could be just updating that game page with data but its way to fast cause time is same for each of them there should be delayed if you sending to page

[dense cosmos]

😭 why

[dry furnace]

explain

[dense cosmos]

finally! the issue was that I forgot to create a move constructor

now I need to get the spacing right

[fast pagoda]

me irl

[dense cosmos]