JPEG XL

```cpp #include <stdlib.h> #include <iostream> size_t ONE_GB = 1000*1000*1000; size_t BUFFERS = 200; int main() { printf("allocating\n"); char* buf[BUFFERS]; for(int x = 0; x < BUFFERS; x++) { buf[x] = (char*)malloc(ONE_GB); } printf("done allocating\n"); printf("checking buffers\n"); for(int x = 0; x < BUFFERS; x+=64) { printf("%d: %p\n",x,buf[x]); } for(int x = 0; x < BUFFERS; x++) { printf("writing %d\n",x); memset(buf[x],69,ONE_GB); } while(true) {} } ```

if you're interested, try using strace and see what system calls are being made during your malloc calls

and maybe check dmesg after crashing

clang may have different behavior

you can do that with strace

according to strace, it looks like it's a lot of calls to mmap and then a sigsegv

at what point are you getting sigsegv? it's probably null pointer dereference because one of your malloc failed

where as TheAznCoder has all successful mallocs but hits oom while paging in pages for memset

No but I’m not actually OOMing as it’s all in compressed memory

I think there’s probably a hard limit set in software

At least in my machine

yeah like oom in the sense that the os is killing your program because of the memory usage

that's the only place it could be

malloc can't crash on sigsegv

glibc bug notwithstanding, ofc

Ngl, I'm tempted to write a library called eral, so the library name would be liberal.so, and conservatives get mad

this is chaotic evil but it sounds fun

Siggreedyprocess

finally, there was time to download all the metadata of art from ||rule34xxx || to create a giant "dependency" graph that does not fit into the RAM (the result of working with tags from 20 thousand art is on the screen)

A porn galaxy!

well, the graph is not particularly useful. it is too obvious that the sound and video tags will stand close.... now I'm going to write an algorithm that will sort all artists by tags on their art. and this will be a useful graph, the similarity of artists in what they draw

Searching for each of the 19 thousand nicknames in 2 GB of RAM TAGS is a very long time (there are still 8 hours left) - I even rewrote the code in JAVA instead of Python - it accelerated code dozen times. Here is a picture of the distribution of the first nicknames in alphabetical order by tags (5% of all nicknames)

graphviz dot 170 megabyte file is ready

😅

I know of only one software (which is easy to use) that can work with graphs of this size, it is Gephi

well... 🧐 it's really a problem to display it all...

The garbage has been removed 2,700 artists are sorting based on 500 tags r34map incoming

funny view of atrists collaboration

<:monkaMega:809252622900789269>

It's really something gigantic... I plan to make JPEG 16k x 16k - the maximum possible for viewing in a browser

packaging problem has been solved

now matrix

Moving to Java to use less RAM <:Thonk:805904896879493180>

more memory was used... But in general, everything fit into 4 gigabytes

the main thing is that I did it today, not in a couple of days

How did you not get rate limited? <:Thonk:805904896879493180>

i do lot of limits. 1. at least 100 different tags per artist 2. at least 100 requests to the tag from 100 different artists 3. Each artist has a maximum of 100 tags (top100) to linking 700 thousand (10% of rule 34 xxx) artworks with a rating above 100 (for 2023-2024, and more than 500 in the entire history of rule 34 xxx) FINALLY I got 1.5k artists sorted by 0.5k tags

full done

https://sami.boo/post/statistics/population-level-differences/

Java should be more performant than python

https://martinheinz.dev/blog/109 `git sparse-checkout` finally solves the problem of downloading just one subdirectory from a huge repo...

https://www.hdblog.it/mercato/articoli/n579286/telegram-900-milioni-utenti/

Also I did Berlusconi birthday in matteo Salvini meta in av01

I was able to do

Probably not legal

But my Facebook triggered an av1 encode

My work - Translate from google. How to properly slow down or speed up a music track: This is done by shifting semitones or tones to preserve musical harmonics . You can't just speed up/slow down to any frequency and get good sound . This is done according to the formula: `Frequency × (2^(semitone ÷ 12 )) Frequency * (2 ** (semitone / 12))` 44100 hertz +2 semitones = 49500 This method uses "modern" music standards of 7 notes by 12 semitones or 7/12 After studying the topic, I found out that this method has an error for the sake of simplifying musical instruments and working with music in general. But in our time this means that the error in octave frequency shift is ~0.2%. It should be completely unnoticeable, but I got to the bottom of it and wrote a program to calculate the best ratio of fractions to approach an irrational number. As soon as I reached the limit in INT64, I entered the last best number into the search and - all this has already been calculated, but who knows where. https://en.xen.wiki/w/User:Xenoindex/Extended_Approximations_of_3/2 And instead of K=7 N=12 I took the ratios K=22885686008 N=39123338641 Next you need to change the formula: `Frequency × (2^(semitone × K ÷ 7 ÷ N)) Frequency * (2 ** (semitone * K / 7 / N))` Here are the pre-calculated frequencies to which you need to slow down or speed up the sound: For 44100 34980 -4 semitones 37066 -3 39276 -2 41618 -1 46730 +1 49516 +2 52470 +3 55598 +4 (Nightcore) For 48000 34979 -4 40344 -3 42750 -2 45298 -1 50862 +1 53896 +2 57110 +3 60516 +4 Ideal shift frequencies will always be fractional. But to get the perfect paranoia-placebo shift, you need to resample the music to a frequency of 384,000 hertz, calculate using the formula how many tones you need to shift relative to this frequency. Then slow down/speed up to the frequency given by the formula, rounding to an even number. And resample the resulting result using the SoX library back to 44.1/48

GENERAL LINKS https://en.wikipedia.org/wiki/Transposition_(music) https://en.wikipedia.org/wiki/Equal_temperament Otner https://en.wikipedia.org/wiki/Semitone https://en.wikipedia.org/wiki/Major_second https://en.wikipedia.org/wiki/Harmonic http://mathemlib.ru/books/item/f00/s00/z0000011/st012.shtml

<@224363555074342912>

I have some original content myself

Proof that it wasn't true that my content aren't wrong value but probably not useful to most people

circular variant of the porn map

spot your guys' fav thing

wait

i thought this was a different server ngl 💀

if anyone wants to know the joke was league

<:PepeOK:805388754545934396> <:JXL:805850130203934781> we all know what JXL is created for - to save even more porn on our computers

it may be a consequence of it

I already have my art folder converted to JXL (although it's SFW art)

I made remake of the meme

wow, someone on the internet just said: > All of your points are correct, my apologies.

you don’t see that every day

Admitting you're wrong is hard, admitting someone else is right is even harder :P

"It's much easier to forgive someone for being wrong than it is to forgive someone for being right."

There are two kinds of people who say wrong stuff on the internet. The first kind is people just pulling crap out of their ass. The second kind is people who want to learn so they pull something crap out of their ass. That way someone comes along and corrects it for then. its free research.

eh, sometimes people are just genuinely misinformed

have you ever heard about that time ford, the auto manufacturer, covered up the ford pinto fire issue because it was cheaper to pay for wrongful death lawsuits than it was to recall the car? it's a commonly cited thing, but it never actually happened

and I'm pretty sure most people who say so believe it did happen because they were misinformed

for example, there was no "internal memo," it was a memo submitted to the federal regulation authority

and was public record

it used assumptions and procedures that were standard and ordinary at the time it was published

and well within the government regulations

https://www.facebook.com/share/v/qivgMwJSUm9ikrFN/

Luca Parmitano how to wash hair in the space

I had tuned gg sans font to jon sneyers standard

you're right

https://itsfoss.com/upt/

Something something 15 competing standards

tho it's a bit weird that they mentiom yum

since yum is deprecated. fedora uses dnf

I mean, they also call Arch’s package manager “packman”

to which an acquaintance of mine reacted with:

lmao

also ```irc <zopieux> nix-env on NixOS, <zopieux> non that's not how it works stop it get help ```

"CPP" environment variable

generally speaking this is an impossible problem though

as not every package manager has the same functionality

or package splitting, not just naming

e.g. on Arch, splitting libraries into runtime vs. dev is the exception (boost) rather than the norm

for example, `pacman -Rn` has no equivalent in dnf

also `dnf remove` automatically does a cascade remove. it removes a package and everything that depends on it

`pacman -R` won't do a cascade remove unless you do `pacman -Rc`

it will just refuse and fail if something depends

What is the difference between this and the GG Sans you've sent earlier?

I'm already at v45

V25 should be old

Weather for wb

Basically loudness regulate for autism

Anyway probably YT will switch to lyra, tabula by jyrki alajyulal

But at least loudness now on ytm is controlled

I've done so many work for many fonts

Open sans open dyslexic gg sans my font 3 aka suduwe

Jetbrains mono

Open office mono tracked

Enwallowify

Use cufonfonts site

Enjoy work has finished

what

<:KittyThink:1126564678835904553>

Out of curiosity downloaded these archives, and all "updated" files inside are dated 2022-2023. So looks like this and all other "work" exist only in his imagination

This is all 2024 made 8:08PM 20march

One this isn't for sure those aren't the originals by Steven Matteson by anyway

But an internet copy with some ascii internet metadata downloaded if you have only internet usage

Gg sans is paid not open source

Irregular reminder that C is a great language ```c #include <stdio.h> int main() { switch(1) { if (0) { default: printf("this is valid and gets printed"); } } } ```

in C, don’t forget the `return 0;`

or have they made it implicit in C too?

In both C and C++, `main`, and just main, implicitly returns 0 if it ends without a return statement.

I thought that was just C++

I thought it was C too

random stackoverflow answer: > Yes, as of C99, reaching the } at the end of main returns 0 if the return type of main is compatible with int.

also I believe that if you omit the return in non-main functions, it isn't UB as long as no caller tries to read the return value :)

> In C90, `main()` must have an explicit return statement at the end to avoid undefined behaviour. In C99 and newer, you may omit the return statement from `main()`.

so they did end up bringing it to C which didn’t originally have it

indeed: ```c long long n,u,m,b;main(e,r)char **r;{f\ or(;n++||(e=getchar()|32)>=0;b="ynwtsflrabg"[n%=11]-e?b:b*8+ n)for(r=b%64-25;e<47&&b;b/=8)for(n=19;n;n["1+DIY/.K430x9\ G(kC["]-42&255^b||(m+=n>15?n:n>9?m%u*~-u:~(int)r?n+ !(int)r*16:n*16,b=0))u=1ll<<6177%n--*4;printf("%llx\n",m);} ```

C99 specifies that reaching the end of main, and only main, includes an implicit return 0

apparently the reason for this is that bjarn strousoup forgot the return statements at the bottom of a few of his main functions in the C++ book

when it was pointing out that his book was noncompliant, he changed the spec, rather than issue a 2nd edition

> It turns out this is all my fault. > > Back in the early 90s, I was exchanging emails with Stroustrup over errata of his C++ Programming Language book. I had found some new ones not currently on his list including missing return 0; statements at the end of some of his example code that used main(). He thanked me and said he would fix those examples. > > He did what he said. But rather than correcting the code, he submitted a proposal to the C++ Standards committee to make his examples legal C++! > > The proposal was accepted and from there copied to the C Standard.

what a horror!

source: a random guy on reddit who claims it's his fault

https://www.reddit.com/r/C_Programming/comments/1ahnkjp/comment/koqk27q/

I didn't want to make it seem more reliable than it sounds

sounds like us with the jxl spec

woah

That assumes most of us get to decide

Doesn't have "I don't take one" :p

I guess I could change this: ``` bindsym Print exec grim - | tee $screenshot | wl-copy ``` to uhhh something like this? ``` bindsym Print exec grim - | tee >(cjxl -d 0 -e 1 /dev/stdin $screenshot) | wl-copy ```

depends on whether screenshots are frames from videos I watch in mpv by pushing `s`

or if they are desktop screenshots

9/10 when I'm taking a screenshot it's to share on Discord so png is the only choice that makes sense

actually, I think Discord properly supports webp on all platforms now? So I guess I could use lossless webp.

me and another friend have been using WebPs on Discord for quite a while now

only thing that doesn't work is animated WebPs because Discord explicitly disallows them and forces users to either use GIFs or pay money to use APNGs

Locking features just so we have to use more room on their servers

APNG is for stickers specifically, not embedded attachments, right?

yeah

remember when they added animated emoji but didn't have a size limit

I have the entire bee movie as a 4x8 grid of animated emoji

they also used to let you upload GIF emoji with a second empty frame and it would go in the animated slot, for servers that were out of regular emoji

you're weird

what does this actually do???

it looks just like a png when i copy it and use it lol

`grim` takes a screenshot, and in this case, sends it to stdout. It outputs it as PNG. `tee` copies a stream to stdout and to a file of your choosing. stdout leads to `wl-copy` which just puts the PNG in your clipboard. The file tee sends to, in the first snippet, is just a name in my screenshots folder with the current date and time, so the PNG is stored there too. In the second snippet, I use `>()` notation which creates a named pipe somewhere, and then gives the path of the named pipe to the application, and connects it as the stdin of whatever program is on the inside of the parens. Which as you can see stores the image as a JXL in my screenshots directory. In both cases the clipboard ends up with a PNG in it.

didn't mean all of it lol

mb for not being more specific

one thing that kinda annoys me is that software that supports all kinds of image formats (like GIMP or Discord) and also support pasting image data from the clipboard, don't support pasting anything other than PNG.

can you embed jxl if it's from a website?

what do you mean?

it i upload a jxl to a website can it embed into discord?

only if the website re-encodes the image for preview

so not jxl exclusively?

what formats does discord embed then?

Presumably PNG, GIF, JPEG, and WebP.

what is <@902036286074925096> for tho?

does it not embed jxl?

It does what I said, re-encodes JXL images into something Discord supports

what a scam lol

thanks anyways

how is it a scam?

what

nvm

Can also do <https://embed.moe/https://jpegxl.info/logo.jxl>

I mean, while we'd love to, we can't use a bot to rewrite the code in Discord quite yet

I understood why android smartphones started taking JPEG screenshots instead of PNG - when opening the gallery, it is extremely difficult to instantly decode 16xPNG ~1080x1920 images. unlike Jpeg

this was not a problem as long as the screen dimensions in pixels were 480p

I remember 8 years ago "they" said that Jpeg was about to die..... but given the modernity, it will remain the most used format for a long time to come.

I realised by default my phone uses jpeg, but if I take an area screenshot then it saves as PNG, even with transparency if it's an irregular shape

hey, this looks kind of <#794206087879852106>

too late 🤷‍♀️

My 5 years old smartphone doesn't have issues almost instantly loading 20 1440x3040 PNG screenshots in the gallery app <:KekDog:884736660376535040>

is it really loading all of them at full res?

keep in mind most gallery apps will thumbnail pictures so you need to clean the cache to test it

you then also have to keep in mind parallel vs sequential decoding, but probably any somewhat modern phone wouldnt have too much of an issue, my s9+ to sequentually decode 20 pngs of 7432x3877 (same png 20 times which isnt super fair because of caching and what not but good enough) has about 2s per png for a total of 44 seconds. that being said there is plenty of left over memory and cpu power so you can easily parallize some of that

my face, when anyone tell me about my CPU 14 years old

i was coding in C++ and wondering why performance SUCKS, then i realised I had NO OPTIMISATIONS ON, turning on -02 sped up code by 35x

does anyone know the algorithm for converting and displaying images with a number (with an error)? it looks like a hash function but for a "floating" image in fact, there are no sources or documentation of the main image recognition services Google Yandex TinEye saucenao therefore, if there is no such technology, I would like to make an algorithm for converting an image into a number, id, hash. to do image recognition by similarity, not on the basis of the images themselves, but on the basis of a number. i.e., image recognition by similarity whitout of the images 🥴

TIL that OpenZFS's kernel module contains a Lua interpreter

Jpeg

Xl or 1

both are preferrable to avif or heic

perceptual hash?

Yes. Absolutely. everything is well thought out.It was invented. in 1980. I'll read about it later, but the pictures show that it doesn't work exactly the way I want it to.

I think heic might be slightly better than avif technically, but it is a patent mess so I would choose avif. But both are better as video codecs than as still image codecs.

*asking on JXL server if JXL is best* Surely it's a little bit biased <:KekDog:805390049033191445>

of course ! the poll was a joke

and how do you do that ? 0.o

nevermind I just found

that's quite a nice feature

at the moment I'm taking the sum of 31 vertical lines and the sum of 17 horizontal lines

I recently downloaded half million artworks from rule34xxx - it was fun. but these are just thumbnails

**fuck cloudflare **i created index.html with all the links to the images and downloaded 50 thousand images per page use chrome

It is absolutely impossible to use any software, curl not work, wget not work, paid BIG soft like OfflineExplorer and other more you can't even execute JS in the browser for massive download (you can not download external links of images in 2024 automatically in the browser)

So I took my hands in my hands

so well...

so I wrote a very primitive algorithm for recognizing identical images. (Perceptual hashing) Stability: maximum compression (like jpeg q0) resizing changing the aspect ratio mirroring in all directions not strong stability: color correction brightness contrast does not work if the image was: cropped (or expanded) rotated

I've read a lot about Perceptual hashing,but for some reason all uses it - splitting images into square blocks. I do the lines, vertical and horizontal. so I got a unique image ID of 48 bytes

```9e5b813939210413656000699581ebd8fff8f7ece1e4d7d0bd7d1c1500141a1a23324b5a574c5f4080bed4ecfffaf2f8```

the first 17 bytes are horizontal lines. technically, its can be used without the entire hash. it works too. but accuracy suffers.

<@226977230121598977> `--disable-features=SplitCacheByNetworkIsolationKey` might help with downloading in chrome since it tries to make it so that images won't get downloaded a second time if you try downloading them with JS or an extension

the images is not loaded at all, even for the first time. instead, image open in a new tab

gallery-dl

You won't even be able to access the site.

I just tested it

how what? ripping from rule34xxx? I use https://github.com/FerrahWolfeh/imageboard-downloader-rs

most booru sites don't care as long as you dont pound them by downloading a large amount of images with massive parallelization

download 1000 images using `imageboard_downloader search -i rule34 -d 10 -l 1000 -o ./dls --id -- "-animated"` to a folder called dls with 10 parallel downloads excluding anything with the `animated` tag, you actually have to patch the app if you want more then 1k downloads at a time though, not sure why, but it is what it is

creating a database for half a million images takes half an hour, the search takes about 5 seconds (on my processor with 1 thread), it art was uploaded a few years after the first one (although I did a test for the noise similarity of the results, and it worked perfectly)

does anyone know how to access all the previews that exist in Google images? 🙂

With how often porn comes up as a topic for image compression, I'm surprised we don't have a PG13 emoji or similar

<:PepeOK:805388754545934396> just me one

today I will be writing a jpeg decoder. to decode 8x8 blocks without reading the information further. because my plans are to create a database of all images on the Internet, and I have created an algorithm that does not require the presence of the entire image. I need to download only the blocks color of images, as I understand ~~1 kilobyte. only Y luma

I will send a description of the algorithm for creating an image hash today in instruct image

I haven't come up with a better name, so the algorithm is called "31+17 id system"

it's good that there are enough JPEG implementations on the Internet, so I just took one of it's and closed the decoding after reading the first values of the block... It always seemed to me. more precisely, I always thought so!... that the first values of the JPEG baseline blocks are always at the beginning of the image (not only in images with progressive decoding). But it turned out that this was not the case. Well, then the whole idea is pointless. Although it was funny

Interesting hash function — have you compared it to other perceptual hashing methods?

it looks like you're breaking a hex hash into bytes and then using a difference of bytes

why not just take xor, and then count the bits in the result?

this way, you end up with a bit in the lowest order byte of the hash contributing 1, whereas the bit in the higher order part of the byte contributing for 128

because this is the Value of the line, and not a separate independent value

oh, then how'd you get a hexdump in the first place?

Yes. I looked at more than a dozen implementations, as well as the official OpenCV docks with Perceptual hashing. -- all bullshit

cause you're doing `"".join(hex())` and then later parsing it back

why not just, pass the list in the first place?

I made a hash visualization

as far as I understand a "percptual hash" is supposed to produce a similar output for a similar input that is only off by a few pixels

the intersection of the lines of segments 17 and 31 (shown with smoothing the boundaries of the lines)

i.e. it's designed to be locally stable

whereas a regular hash is by design locally chaotic

I have already assembled a team of 5 people to parse the Internet, as well as create a search site (I will do it myself)

yep

it would be nice if there are interested people here to parse, for example, reddit

since the created hashes (IDs) are independent, it is easy to simply add them to the database like HASH:URL

Prince of Persia 2 (1993) had some catchy music

(in context: https://youtu.be/idGpzB0g4wA?t=32m20s )

For some reason that reminded me, my MIDI emulation in Windows is horribly broken for some reason

Sounds like I have a drunk band playing in my game

I'm partial to the 3D Ultra Pinball 2: Creepnight Soundtrack

they really captured the halloween aesthetic

for this kind of thing, I think I’d tend to gravitate towards some of what Alberto Gonzales did for The Smurfs’ Nightmare https://youtu.be/2E7M0fUDcnU?list=PLDL64W9EXILGr59rWaLqm34Wn4yPLZwdJ

(track 5, “Hidden Around”, may be a bit more Halloween-like)

(updated the link to a higher-quality version)

this one is also an impressive piece: https://youtu.be/ZXlpzbIl1Bg

https://youtu.be/d6iQrh2TK98?si=uOp2oqRd_8hR1t18

is there a release or binary or something for the *original* ssimulacra? I'm having opencv config issues that I don't want to fix 🙂

https://github.com/cloudinary/ssimulacra

does this fix them? https://aur.archlinux.org/cgit/aur.git/tree/opencv4.patch?h=ssimulacra

yes, thanks a bunch

I think Fab is going a little overboard again lately...

and this is why I have him blocked

he was actually kicked out from the av1 server for the constant spam and pinging

also you good sir have my thanks

mods don't care quality of the server plummeting

https://youtu.be/XkSrlZLtTbE

any good VPS recommendations, preferably with locations in Asia for latency purposes? i'd try Hetzner, but they don't have Asia locations idk if i can trust Vultr after the whole ToS fiasco Linode rejects my account immediately for some reason... and idk if i trust those random small VPS providers, i want something a bit more reliable and hopefully with decent CS, what's your experiences?

<@693503208726986763> thank you for infecting me with the urge to mega optimize jpegs and turning my PC into a space heater <:KekDog:884736660376535040>

I'm honored to have pleased you

https://cdn.discordapp.com/emojis/586100936544616450.png?quality=lossless&size=48

<:KittyOwO:1147753615054872597>

how to solve this?^

ok bye,

note that I use `-i` with that tool even if it says `Allow multiple planes per DC scan, which may improve compression (incompatible with some software)` `-i generates files that are incompatible with some software such as Photoshop and Opera <= 11.61.` I've not seen any app not supporting the output, and it compresses even better some files with that flag

Same here, but haven't tried if it breaks in Photoshop

I've had many issues with Photoshop with optimized jpegs in the past, so wouldn't be surprised if it breaks compatibility

0.o

Wonder how much of a difference using jpegtran vs mozjpegtran makes <:thinkies:854271204411572236>

``Optimal compression is achieved by specifying the mozjpeg version of jpegtran for the -o switch. However, mozjpeg is slower and buggy with certain scan configurations, so IJG jpegtran is recommended for the -p switch.``

jpeglitran when

I got once a file which I got an error with: 'too many scans' or something like that I wonder if IJT would have worked, but was a bit lazy lol

ah shit

I sent one big message

while I could

just send

several

small

messages

to reach faster level 10

:trollhq:

Don't go fab mode <:RaysShock:686219918030798921>

Sadly that's not how it works lmao

Think there's a 1 or 2 minutes cooldown

seriously ? shit <:KekDog:805390049033191445>

Why am I even doing this

Have been optimizing a few large images for the past few hours lmao

Expected it to take long, but not this long <:KekDog:884736660376535040>

exhaustive bruteforce 💯

The only way to do it <:KittyUwU:1147753612529913938>

if someone else if paying the electric bill...

The only problem is that I can't really use my PC while it's optimizing images <:KekDog:884736660376535040>

Except for watching videos I guess

just don't use all your cpu cores for optimizing JPGs lol

you're on Linux ?

Nope™️

<:KekDog:805390049033191445>

I could just use less threads, but don't want to slow it down

on Linux I know how you could restrict cpu usage of an already-launched process

Hm, using mozjpegtran made a decent difference

Oh wait, nevermind. Input was the non optimized file <:KekDog:884736660376535040>

Either something went wrong or it just didn't matter at all since the output was the same XD

easy

.

In fact, I haven't written for a long time. but I have already typed a million images into the database.... and I no longer have time to working with it. Because I bought a ryzen 5 7600 <:H265_HEVC:805856045347242016> <:AV1:805851461774475316> <:H264_AVC:805854162079842314> <:JPEG_XL:805860709039865937> and compress time

Attention breaking chBnges libjxl encoder

I will just continue how much i would like to wanting i don't warn you

I think my changes are benifical for the oxygen of 99% neurotypical people and there are studies confirming that and conforming that

So I don't question it

Is a good change

This likely would be allowed even in jpeg xl and people would have what they wanted and be fuery i don't care about other people priority cause i prioritize only me i'm autistic and full racist

I'm autistic type 3 and i want this on the internet to happen than the others talking about orders especially ryssians and xrhinesse

To be honest Id don't know anything about jxl specifications but only that

https://www.instagram.com/p/C5f8hT0tlev/?igsh=MTNhZzg0aWpyaTdqeQ==

Yeah, it's starting to become a bit... much. I think the spam increased after he got banned from the AV1 Discord server for off-topic spamming everywhere and it's decreasing the quality and readability of this server by a decent amount :/

Yeah you right he's a chinese flat encoded with libjxl 0.10.0

Not perfect but I've never done those type of images in past 10 days

@fab I am muting you for a week, you are spamming incomprehensible messages all over all channels to the point that it becomes hard to follow the actual conversations. If it restarts like this next week, the next step will be a ban.

It's impossible to follow what he's trying to say too. I really think he's a malfunctioning language model

new personal record: 25781 tabs open

<:monkaMega:809252622900789269>

currently have 912 tabs, but it took me 2 years lol

how can you reach 26k <:KekDog:805390049033191445>

opening tabs

not closing them

oh wait did you mean like, benchmarking a browser to see at how many tabs it crashes ?

Maybe I should take back that I called someone chaotic for having 30 tabs open <:KekDog:884736660376535040>

on my phone tho I have only a bit more than 100 tabs iirc

I'm somewhere over 12,000 rn although it would have been way higher if I hadn't lost my sessions multiple times in the past

which reminds me of something, Firefox is removing the config option that lets you decide whether or not all tabs start loading when you restore them

I've never had the need to change it from the default of only loading the current tab but still it's lame they are removing the option

no

https://i.imgur.com/IShqU5E.jpg https://i.imgur.com/XhpNeJ8.jpg

quick how much erotica can we find

in an effort to clean things up, I've so far closed 50 windows

this is why I use a tab manager

when I am finally done I can just nuke the tab group im working on

I kinda wanna start using Firefox's old Tab-Candy/Panorama/Tab-Groups system. there's a fork of it here that works with modern Firefox: https://github.com/117649/Tab-Groups

here's a reallllll old video showing it back when it was in development

Tabs stress me out

bro just close them and use the history menu

it also has a search box

history has a size limit

LORD

Yes my child?

https://youtu.be/-G9U1QY4UDs > I am the Lord of Darkness

https://twitter.com/archer_uwu/status/1779035120487313527

I’ve been using the web client, I don’t remember exactly why

I think maybe I was concerned that the native client wouldn’t be colour-managed

no real downside to web honestly

if anything it's better

There is no "native client"

The native client is just a browser

lol

So it's like having 2 separate browsers eating RAM instead of just 1

iirc, when I looked it up, what was said to be the one advantage of the standalone app is the ability to use a global push-to-talk key (instead of only when discord has focus)

which I didn’t care about

actual working audio on linux is nice too.

My friend has to use the website on their work computer and complains about the lack of noise cancellation in calls along with strange extra noise getting added somewhere along the line

I'm archiving youtube videos. Sometimes they are available in multiple formats, like vp9 and av1. Do you know of some heuristic to tell which version has better quality, besides size? I have ~500 videos in a range where size alone is not enough to tell. Maybe some kind of AI? Or some tool to calculate metric based on edge detection, since av1 and vp9 tend to oversmooth files when there's not enough bitrate? Obviously I don't have the original file to compare so it's tricky. Histogram of the size ratios:

Not having a source to compare to makes it pretty hard annoyingly... I suppose one idea would be finding a keyframe, moving 1 frame along and then comparing the difference, since that should be 'clean' and then whatever bitrate they gave it to work with

The keyframes in av1 and vp9 are not the same frames. So do you mean compare frames that are not keyframes in both? Also, once I calculate the difference, how would I, in a somewhat automated way, choose the better version?

actually I have to check if they are the same

So far I'm just thinking out loud, the keyframe idea was from messing with files encoded with only 256kb/s in WebM and seeing the first frame (keyframe) immediately drop down to the bitrate

It is actually the same

It would be interesting to try to come up with a way to combine several recompressed videos (that were presumably all encoded from the same original) and to make a best guess of what the original looks like. Surely there should in principle be more information available if you combine all sources than if you pick one.

I am keeping one channel (Unus Annus) in both formats specifically in hopes something like that happens is the future

Good call

especially since it's been deleted and I don't think the av1 videos are accessible anywhere. I should probably upload them to archive.org

Vp9 is not quality if you look for 10000+ videos disfare vp009

Bitmovin explains so well

https://bitmovin.com/av1-4k-video-sd-bitrates

https://gizmodo.com/the-quiet-danger-of-noise-canceling-headphones-1851407784

unfortunately, it seems to have resumed

I'm just not smart enough to understand

First time I had to use <:BanHammer:805396864639565834> on someone who is not a crypto-spammer or something like that.

I feel a bit sad but it had to be done.

Yeah, it's unfortunate since he doesn't have bad intentions. Just sad that he doesn't really learn from warnings

IIRC I gave him a one-week timeout five times, each time he just resumed when the timeout was over. I can live with some noisy communication and idiosyncratic behavior, and this is only an informal chat anyway, but the fact that he goes all over the place (in several channels) and tends to post stuff that takes quite a bit of vertical space made it really quite annoying and disruptive.

Not to mention when he posted to every github issue we had been associated with during the first timeout

and the random mentions

That sad moment when a very old jxl version creates a vastly superior result over a new jxl version <:KekDog:884736660376535040>

On the positive side, at least new cjxl was about 6 times faster with encoding XD

Oh, interesting. Discord put the JXL files into the 'gallery view' where you can just click left and right. Doesn't decode naturally, but now they're recognised

still in love with the guitar here

(I’ve tried to play those arpeggi; it’s… not easy)

>We need to give the HTML5 spec to ISO so that the spec can be locked behind a paywall and we can start patenting the spec -Uni club

I've been unconscious between 2024/04/15 and 2024/04/19, hospitalised without access to internet till 2024/04/24. I've just caught up with the @fab lore. He used to be different. A psychiatrist recommended me to take a test for Asperger's. I am. Will take around a year. (side note: I believe ICD-11 (2019/05/25) still hasn't been translated to Polish, even though it's in effect (2022/02/21), and I told him that I thought Asperger's is not used anymore and he told me that it is, but seems both dsm-5 and icd-11 just talk about autism spectrum? so is it??) During my hospital stay I've heard some wails and almost died myself. And now @fab is gone. I am feeling.

correct, Asperger’s syndrome was merged with the rest of autism > Other enhancements have been introduced to promote ease of use across all settings: > […] > - Consolidation of autistic disorder, Asperger's disorder, and pervasive developmental disorder into autism spectrum disorder. > […] > Within the diagnosis of autism spectrum disorder, individual clinical characteristics are noted through the use of specifiers […], as well as specifiers that describe the autistic symptoms […]. These specifiers provide clinicians with an opportunity to individualize the diagnosis and communicate a richer clinical description of the affected individuals. For example, many individuals previously diagnosed with Asperger's disorder would now receive a diagnosis of autism spectrum disorder without language or intellectual impairment

but some people in the autistic community are quite critical of the DSM anyway, and especially of its pathologising language

for example, the author of “Neuroqueer Heresies” has said: > On top of that, most of these descriptions of autism – even many of the descriptions written by autistics – propagate inaccurate information and false stereotypes. Some are so bad that they actually quote the DSM.

I see there's a lot to learn.

/gen

on the frequent desire to “subtype” autism, the authors of “Autistic Masking: Understanding Identity Management and the Role of Stigma” have noted that: > The labelling of autistic people as ‘heterogeneous’ is dehumanising. Whilst people assumed to be neurotypical are labelled as having ‘individual differences’ that impact on their similarity to each other (e.g. differences in attention or memory), autistic people are often assumed to be a monolith, despite early research highlighting the variation across the population (Wing & Gould, 1979). > > The emphasis on similarity, and minimisation of individual uniqueness implicitly suggests that whilst humans in general are complex and differ from each other in many ways, autistic people are expected to show more similarities. and: > Overall, both the diagnostic criteria for autism and the societal understanding of autism is limited, and our current understanding of what autism ‘looks like’ relies both upon stereotype and emerging knowledge about IM [Impression Management]. Researchers have attempted to address these issues by proposing the introduction of phenotypes/subtypes, keeping the core construct of autism intact. However, these divides are based on the assumption that the core construct is actually correct in the first place, and worth preserving.

Note that Hans Asperger collaborated with the Nazis or at least did some dubious things and seemed to hold beliefs supporting Aryan "race hygiene". I prefer not to honor such people by naming something after them.

what about the rest of half of everything discovered and invented

I don't think I will be able fix myself from hearing it as south park's "ass burgers" that and fish sticks

~~Who doesn't like to put them in their mouth <:thinkies:854271204411572236>~~

it reminds me more of asparagus

(all the more because asparagus in French is _asperge_)

https://fr.wikipedia.org/wiki/Asperge

<@703028154431832094> 's av1 for dummies server has a fab channel

<@224363555074342912> thank you, but It's not that I miss the messages I could not understand, I just feel sorry for Fabian who through no fault of his own is going through something like this. I also reflected on how ephemeral and uncertaint life is. (Which I am absolutely willing to admit is a wholly unoriginal thought and just a part of growing up everyone goes through and this week it was my turn)

Yes, that too. Both in French and in Dutch we say "asperge". It's the season by the way, just the other day I cooked "asperges à la flamande" which is probably the most popular way to eat them here. It's very delicious. https://www.biendecheznous.be/recettes/asperges-la-flamande

that with some steamed potatoes (I like small ones, in the skin)

It's hard to estimate what exactly @fab's situation was/is but yes, whatever it was, I also feel sorry for him. I don't think there's an awful lot we could do to help though, but I like to think that perhaps getting banned from this discord will somehow slightly help him to get the change he needs. In any case, I hope he has people in his more direct environment who can help him or find help.

# uh oh. so far, the development of my software is very slow. but I'm doing it anyway. I wrote code to index downloaded groups from **telegram** in JSON format quickly and also wrote 2 implementations for searching images in the database. for online (1 photo) and for offline (as many photos as you want). so far, there are only optimizations for 1 photo(online), if a little, then it's better to look there. Optimizations for multiple searches have been disabled so far. I plan to write a discord parser on the downloaded server with images. because I have no idea where else to get hundreds of thousands of photos to search for them. google does not give its database of Google photos unfortunately *in all seriousness, I use this software only to see which pictures I have already posted* but nevertheless, hundreds of downloaded TG groups is usefulness (some have the author's nickname on Twitter)

the only thing that gives me motivation is the complete absence of recognition collisions on a database of 1 million photos

the database is not private. it just 80 md in 7z and I can only send a small part of it. but it doesn't seem to make much sense. you will definitely not find the photos that you have

The search information, or as I call it IMAGE ID, size only 80 megabytes per 1.5 million images

Had a significant earthquake here again

oof

the idea of writing a blog post about the “sleeping beauty paradox” has occupied my mind for a few days, and I’ve finally started to actually write it

it will obviously settle the question once and for all

the arguments for 1/3 are fallacious in variously subtle ways

So frequently that another just happened

Where?

Nice! I had planned to go to bed, now I have a full wikipedia article to read before

this problem is possibly the worst nerd snipe I have ever come across (for me)

https://xkcd.com/356/

(I am trying to come up with a setup in which 1/3 is the correct answer, and to pinpoint exactly what it is that makes it different)

(I have the setup, I’m now doing the latter)

In Taiwan. Not a famous place

I spent half a day creating this track from 50 attempts to generate different samples. and then I sewed these samples and do mastering, and why no one appreciated (anywhere at all). You can listen to https://youtu.be/gZE4CfIjgg4

No matter how hard I try, I can't figure out whether the neural network is not interesting for anyone, or music in general.

although it is definitely difficult to say whether the soul appeared in the music from the fact that I combined it <:Thonk:805904896879493180>

the more I think about it the more obvious it seems to be it can only be 1/2? even if it was 1 vs 1e6 wakings. I'm looking at Wikipedia and I don't have a precise definition of "degree of belief", but for one, it seems whatever comes up on the coin, the beauty is only being asked once?

if the coin comes up tails, she is asked twice (once on Monday and another time on Tuesday but with zero memory of being asked on Monday and no way to distinguish the two)

I argue that the number of times she is asked shouldn’t affect the answer

ah wait, she is being asked multiple times. but those are not independent variables.

in fact, the problem is exactly equivalent to asking her to write her answer on a piece of paper and reading it once or twice depending on the outcome

no need to put her to sleep at all

even 1 vs n times being asked, it's 1/2, and in half the coin outcomes she would be wrong

Ambiguous question position edit Another approach to the Sleeping Beauty problem is to assert that the problem, as stated, is ambiguous. This view asserts that the thirder and halfer positions are both correct answers, but to different questions.[10][11][12] The key idea is that the question asked of Sleeping Beauty, "what is your credence that the coin came up heads", is ambiguous. The question must be disambiguated based on the particular event whose probability we wish to measure.

That argument that she knows beforehand that the probability is 50% and doesn't ever gain more information is quite compelling...

Wikipedia copy paste

as far as I can tell: it would be 1/3 if the question were “if, in a series of repetitions of this experiment, we pick one awakening at random, what will have been the outcome of the *last* coin flip before that point?”, not because the probability that we assign to any single specific flip is different from 1/2, but because _which_ flip we are talking about is a moving target, and the flip from two awakenings ago can be the last flip if and only if it (there was one and) it came up tails

like if I'm betting money each time I'm being intervied, I'd not say 1/2

indeed, but it’s arguably because the betting procedure itself is distorted

“what will you bet? btw, if it’s tails, the bet will be executed twice”

yeah. I like the ambiguous question answer the most. I was disappointed when the article on degree of belief was so lackluster

as a mathematician I consider the solution to be fairly obvious

if she knows how the experiment works, then she knows the probability is 1/2

there's not much else to it

the author of the problem does offer an argument for 1/3 (https://www.princeton.edu/~adame/papers/sleeping/sleeping.pdf), but the argument is flawed (as could be expected)

at a meta level, P(the answer is 1/2 | that argument) = P(the answer is 1/2) × P(that argument | the answer is 1/2) / P(that argument)

which is still high

and therefore, the probability that the argument is wrong, also high

the thing is that the exact way that it’s wrong can be a bit subtle; hence my attempt at exposing it for all to see

I think I’m almost done, btw

I’m proofreading and checking whether there are any obvious reinforcements to make

let us know if you want an external proofreading

😄

is it math monday ?

here we go: https://sami.boo/post/statistics/sleeping-beauty/

1: What the fuck does this mean

2: Does this even happen? What is the point of such an experiment? Why are we thinking about a scenario that just won't happen?

``` 2022-04-12 13:32:27 spider-mario oui mais ça arrivera pas donc à quoi bon y penser 2022-04-12 13:32:29 spider-mario problem solved ```

https://tenor.com/view/shocked-face-wide-eyes-bug-eyes-shocked-surprised-gif-17836791

https://youtu.be/Rr_T4Aim6Fw

this is unironically the best abuse of notation I've ever seen

rewriting it as `(1 - int) f` where `1 - int` is a differential operator legit works

like it's a stupid meme but this unironically actually checks out

though you have to make some magic

basically `int` is a differential operator which is a linear transformation from `C^inf -> C^inf`

and instead of "dividing by both sides" you're taking an inverse differential operator

since the kernel is 1-dimentional, the "inverse" is actually a family of functions spanned by a single vector, and that's where C comes from

as written, it's an abuse of notation

but it can be made rigorous using Linear Algebra