[ home / bans / all ] [ amv / jp / spg ] [ maho ] [ f / ec ] [ qa / b / poll ] [ tv / bann ] [ toggle-new ]

/maho/ - Magical Circuitboards

Advanced technology is indistinguishable from magic

New Thread

Options
Comment
File
Whitelist Token
Spoiler
Password (For file deletion.)
Markup tags exist for bold, itallics, header, spoiler etc. as listed in " [options] > View Formatting "



[Refresh] [Bottom] [Catalog] [Archive]

File:1000003087.jpg (90.18 KB,850x1133)

 No.2865[Reply]

I'm trying to figure out how the youtube algorithm works when logged out.

When I visit the site from a new private window and look something up, the results seem to be pretty consistent across different browsers and sessions. But if I visit the same content as linked by a website, I get completely different (much worse) stuff, and it confuses me. Does it really put that much stock into that initial search?

I also just kind of wonder what youtube looks like to people just getting into it now. I made my account over a decade ago, so I have no clue where things are at with that kind of thing.

 No.2866

I really don't know how it works, but I haven't been logged into youtube for at least a decade. Sometimes it recommends some neat things, but it likes to throw curveballs at me sometimes. Sometimes I'm watching it in bed and I fall asleep and wake up to conspiracy theory stuff because I was watching history. Kind of annoying.
Opening youtube in a private tab to see the base experience is about what you'd expect- just normal people stuff like MrBeast or world news and sometimes surface-level 2D stuff like whatever the big anime of the moment happens to be. The good news is that at least at the beginning it doesn't feed you outrage content, but maybe that's it being purposely deceptive.

 No.2874

I think if they can place a backlink from where you came from and then have identifying traits of people coming from that backlink then they can knbow what to recommend to you from there. Which would explain your results.




File:1513485451722.jpg (66.72 KB,736x588)

 No.2420[Reply]

https://github.com/erengy

Is it possible to use the repos here to reprogram and create a Taiga that works on linux? I'm asking because I don't think I can possibly go back to keeping up with anime using just a notepad anymore after how convenient Taiga's been for me this whole time, and I'm not switching to Win 11 or the subscription-based 12 in the future. If I need to learn to code myself to do this then so be it, I want my convenience.
13 posts and 2 image replies omitted. Click reply to view.

 No.2435

File:[A-FanRips]_Roujin_Z_[BD.1….jpg (354.02 KB,1916x1076)

>>2434
Uh, hm. I'm not sure actually. Didn't really know of its existence, but it tracks manga/vns as well? That's pretty neat.

 No.2436

File:Smug_Ayanami_2.jpg (731.98 KB,1920x1080)

>>2434
>trackma
trackma nuts

 No.2439

>>2420
>Win 11 or the subscription-based 12 in the future.
omfg what

 No.2869

File:1744685819151.gif (541.54 KB,212x219)

I just watch 1 anime at a time. I dont watch anime thats still airing. no need for fancy taigas and stuffs

 No.2871

>>2436
Good one anon!




 No.2144[Reply]

Have you ever wondered why all the AI coding demos always use things like "Create the game of snake" or "Create a todo app"?

 No.2145

no, seems obvious to me

 No.2146

we need another tinder clone

 No.2147

twitch plays tinder

 No.2867

>>2144
Because they're dead simple programs and can be one shot easily




File:genius_comes_17.jpg (768.16 KB,1064x1500)

 No.113[Reply][Last50 Posts]

Today I begin a programming project to construct real numbers and various operations on them. I don't mean the wimpy limited precision floating point numbers that you've probably used if you've ever written a program in your life. Nor do I mean arbitrary precision numbers which can have as many digits as you like (until your computer runs out of memory) but still have irrecoverable rounding error because the digits end at some point. I mean exact representations of numbers that can be queried to any precision (which in practice may be limited by time and memory constraints).

This has been done many times before and probably much better than I will manage, so I don't expect to create anything really novel or useful. This is primarily for my own edification, to learn more about numbers and the computable/constructive reals. In this thread I will blog about what I'm doing and discuss the philosophy of numbers with anyone who's interested.
203 posts and 61 image replies omitted. Click reply to view.

 No.317

Correction: There are some < signs in >>316 that should be ≤ signs.
Instead of
> 2 < (r/s)^2 < (√2 + ε)^2
> 2 < r^2/s^2 < 2 + 2ε√2 + ε^2
> 0 < (r^2-2s^2)/s^2 < 2ε√2 + ε^2 = ε(2√2 + ε) < 3ε < every positive rational.
it should say
2 < (r/s)^2 ≤ (√2 + ε)^2
2 < r^2/s^2 ≤ 2 + 2ε√2 + ε^2
0 < (r^2-2s^2)/s^2 ≤ 2ε√2 + ε^2 = ε(2√2 + ε) < 3ε < every positive rational.

 No.318

>>315
>>316
So to answer the question of whether there's a number system with a number ε such that
¥ 0 < ε < 1
¥ 0 < ε < 0.1
¥ 0 < ε < 0.01
¥ 0 < ε < 0.001,
and so on, the answer is yes, there are many. There are the hyperreal numbers which are useful as an alternate way to do calculus. There are the surreal numbers which arose out of game theory. And you can even just postulate such a number ε and add it to the reals sort of like how we add i to make the complex numbers. In the case of the complex numbers, you get the complex numbers from all the linear functions of i. If you add ε, your new set of numbers consists of all the rational functions (polynomial divided by polynomial) of ε.

But we can see that a nonzero infinitesimal number like ε will cause trouble if we want our numbers to be expressed by infinite decimals. The whole idea behind infinite decimals is that we can identify a number by comparing it with finite decimals. If we allow ε into our system, then we can no longer distinguish ε from 2ε, or 0.4 ÷ 0.03 from 0.4 ÷ 0.03 + ε, or √2 from √2 + ε by comparing them with finite decimals. We would need something else. For example, in the surreal numbers, ε is the simplest number between {0} and {1, 1/2, 1/4, 1/8, ...}, and 2ε is the simplest number between {ε} and {1, 1/2, 1/4, 1/8, ...}.

Another definition:
¥ Two numbers are infinitesimally close if the difference between them is infinitesimal.

There are many equivalent ways to construct the real numbers, but all of them are designed so that we never construct a real number infinitesimally close to another real number. I don't think anyone knows whether it's possible for two real-life quantities (such as distances, times, weights, and volumes) to be infinitesimally close to each other yet not equal. If it is possible, then when we model these infinitesimally close quantities with real numbers, we are choosing to ignore any infinitesimal difference and assign the same real number to both.
Post too long. Click here to view the full text.

 No.319

>>312
>They made use of a preexisting Java library to do the calculations:
>https://www.hboehm.info/crcalc/

I said that like they just used someone else's code, but on second look I see the guy who wrote that library is the author of the CACM paper. He developed it while he was working at SGI and then got hired by Google to work on Android stuff.

 No.320

File:infinite decimal division.jpg (209.64 KB,724x1209)

>>308
>>311
I've been thinking again about how to do hand arithmetic with infinite decimals, in this case division. I've tried this before, but my calculation of the error on the result at each step was an ugly and unmemorable formula. I think I've found a better way to think about the error on the result, which is to first work out the error on the remainder at each step. There are two contributions, one from the truncation error on the dividend, and one from the error on what's been subtracted from the dividend (due to truncation error on the divisor). Then the error bound on the result is
(|remainder| + |error on remainder|) / (lower bound on |divisor|).
You could also forgo taking the absolute value of the remainder and work out separately the error bounds in each direction, which could occasionally help squeeze out a digit of the quotient a bit earlier.

 No.2863

bump for respect
thanks




File:Screenshot 2025-04-05 1217….png (58.6 KB,403x106)

 No.2829[Reply]

Apparently YouTube is experimenting with using AI to dub videos.

https://www.youtube.com/watch?v=atl5V2Osk80

 No.2830

how do I get yt-dlp to download the dub track?

 No.2831

I recognize that voice from the rainbow kakapo classroom fad from a few years ago.

 No.2861

>>2830
With any luck you can use a flag to list the audio tracks and snoop out the english one.

Done it before but I forget out.




File:926190f09d.png (180.33 KB,1920x1040)

 No.2699[Reply]

I became a Zen supporter.

I like how they improved upon Firefox's tabs system. It was essentially the only thing that prevented me from going to chromium and they made it better
20 posts and 6 image replies omitted. Click reply to view.

 No.2849

>>2843
vertical tabs are better. I like seeing my tabs instead of them shrinking as I open more

 No.2852

Zen does have issues, but it's an Beta product.

 No.2853

I use zen twilight, only issue I've had (that didn't get patched) is that tabs I havent had as an active tab for a while will randomly act like I clicked to close the page with text submitted even if I didn't, and ask if I want to close the page and lose info. Was a little annoying when wiki editing

 No.2856

File:56f52301f5.png (25.34 KB,1230x162)


 No.2857

My idea is to put the tab bar to the right of the address bar.
Just cram it all in one bar, hamburger menus everywhere.




File:Deppseek.png (44.92 KB,750x750)

 No.2334[Reply]

There's been a lot of chatter lately about Deepseek. In the online circles I'm in, people have a politics-colored understanding, more or less saying "American tech companies couldn't do this, but an opensource Chinese company could and American tech companies are in 'damage control'". Which... I really don't understand. If it's an open source model, like Llama was, for example, I don't see how this doesn't just cause there to be a proliferation of much more efficient and performant models -- the same way after Llama became available, sudden there was Phi from Microsoft, Gemma from Google, Mistral, and others.

What does /maho/ think?
65 posts and 20 image replies omitted. Click reply to view.

 No.2523

File:[Erai-raws] Medalist - 06 ….jpg (221.15 KB,1920x1080)

>>2519
I ended up ordering a 2x48 thing, but I feel immensely ripped off by it. This is the computer I'm going to be using for the next 4 or so years so I justified the price premium to myself since I don't buy things like real clothes. The whole reason I'm doing the RAM thing is for local text gen so I'm eager to hear of your progress in it. (saw in IRC that you're having issues though...)

>It depends what you mean by "the great model".
The famous one that is a little bit below GTP4o or whatever. My assumption is that I can't run that one, but maybe the one under it? Slow 1 token/s generation is rough, but part of the appeal of all this VRAM and RAM stuff is multitasking.

 No.2524

The API is still fucked huh? I assume results are slightly worse than usual because of the DDoSing or whatever

 No.2541


 No.2850

>>2334
I think its based they made an open source chatgpt challenger. Open Source is one of those few things I feel wide eyed optimism about its great. (Even though open source devs are often unhinged in my experience)
>Which... I really don't understand. If it's an open source model, like Llama was, for example, I don't see how this doesn't just cause there to be a proliferation of much more efficient and performant models
They don't care about that theyre American corporations that prefer closed proprietary stuff they can sell. They're cold war esque paranoid about the Chinese surpassing America at anything. I watched the news when this was current and there was a swirling torrent of whining from American talking heads
>>2335
I don't think that matters but you could explain really easily.

 No.2854

File:[Erai-raws] Chuuzenji-sens….jpg (220.69 KB,1920x1080)

Heh, interesting bump since I've been looking at local text gen stuff again. Doesn't seem like I would be running anything related to deepseek. For the big deepseek thing you need a server motherboard to fit all the RAM slots and buying all the RAM would be way too expensive for me for a specific purpose like this.
It seems like the local text gen hobbyists are waiting for "QWEN3" to release sometime in the coming weeks, although I'm not sure how good at roleplaying it would be. Seems like the era of random merges to attempt to get good uncensored RP is mostly a thing of the past.
The good news is that the models they run fit inside 24gb of VRAM, although the bad news is I don't have that much since my plan was to get a 5090 and that will simply be an impossibility now.




File:Medalist.S01E01.A.Genius.o….jpg (327.23 KB,1920x1080)

 No.2269[Reply]

I'm feeling really, really tempted to get a 5090 even though it's a massive scam and it's months worth of savings. It's not the rational thing to do, but I do AI stuff a lot and it brings me joy. (no I don't condone AI shitting up the internet and art)
There's also 3D modeling I want to take more seriously after Palworld reignited my passion for building stuff. Obviously you don't actually need a top of the line card to do this stuff, but it does allow more geometry to be active and speeds up rendering massively. More VRAM means you can have more processes open so jumping between programs is smoother.
AAA gaming sucks apart from Capcom so that doesn't really enter the equation at all. I guess ironically retro pixel filters are known to be VERY demanding if I decide to do that. Might be more demanding on CPU, though, I can't remember.
I'm in that CG tracker that went private 5 years ago so software and assets are no issue, but man this is still such a huge amount of money.
What to do...................................................................................................................................................................
74 posts and 38 image replies omitted. Click reply to view.

 No.2819

File:firefox_QrhLTNmL7M.png (518.27 KB,1480x946)


 No.2820

>>2819
>$5500
there are places where you could buy an apartment for the price of a few of these

 No.2821

File:>o>.jpg (127.52 KB,381x374)

>>2819
>5500
That's a little over $1/RTX!

 No.2822

>>2820
That price is actually pretty close what i bought my previous apartment for. Then again, that was like a decade ago and it's probably several times more expensive now, so i guess your point still stands.
Imagine putting "new graphics card" as the reason for a mortgage application...

 No.2828

File:waterfox_3j0NJ5XF4B.png (87.33 KB,1100x259)

These restocks that aren't at 7AM are really my only chance and I missed it. Oh well.




 No.2814[Reply]

https://www.youtube.com/@Firerama

They're making YouTube games with the caption system now

 No.2815

not really a game, just looks like an effect

 No.2816

maybe this means youtube has given up trying to block invidious and yt-dlp, and this is how they're trying to get people to come back to their ad and tracker infested website

 No.2817

> get people to come back
You were never a significant client base




File:Flutter.zip (1.22 MB)

 No.2769[Reply]

I am ADDICTED to this fucking yoshi's island romhack called Flutter
the premise is that to beat the levels 100% you need to use all sorts of edge case interactions between sprites and yoshi transformations and swimming that you would have never come across in the original game
the hack does a fairly good job of teaching you these things but it does assume you know the original game inside out

 No.2770

That reminds me of this old, obscure SMB romhack called AIR which is loosely based on the VN of the same name. You only get one life and can jump in mid-air infinitely but that doesn't make it any easier: you also need to know the game inside out and be skilled at pulling off different glitches to beat it. This is a tool assisted speedrun so obviously it powers through the game but it's a lot harder than what it looks.

 No.2771

ROM hacks that are clearly made by people that really love a game and want to spend more time in are pretty awesome. The most common is probably the 'randomizer' that seems to be hugely successful for streamers and it leads to tournaments and other things which highlight the player's knowledge and ability to adapt to things.
They're cool, but a handcrafted thing will always be better I think. Another cool thing with these rom hacks is if you buy a flash cart thing they're playable on actual hardware. I would have killed for this stuff to be available back in the day when these consoles were new. Imagine staving off boredom with all this stuff!

 No.2810

File:C-1742814324321.png (25.82 KB,512x448)

hard section... here you have to race kamek and prevent him from reaching the left side of the screen, and the bit at the bottom is an obstacle course where you have to use melons and eggs to melt or kill stuff and you have to spit enemies at other enemies and theres no margin of error. the stuff right above yoshi is instakill blocks

 No.2811

the whole romhack is like this
that and a looooot of timing puzzles
timing is a pretty solid thing to build a wide variety of puzzles around because all the different elements you can use in yoshi's island have different delays and speeds. theres a lot of richness




File:rtx 2080.webm (1.41 MB,640x360)

 No.2776[Reply]

https://www.pcgamesn.com/nvidia/rubin-gpu-early-release
https://www.tweaktown.com/news/102949
>NVIDIA's next-gen GeForce RTX 60 rumors begin: Rubin GPU, more VRAM, DLSS 5, TSMC 3nm node
>2026 Nvidia GPU hardware might launch six months early, but don’t panic yet

I've seen this show before...

 No.2777

File:TASOKARE.HOTEL.S01E08.The.….jpg (244.03 KB,1920x1080)

>Rubin’s bump-up would be specifically on an enterprise level.
>While this could potentially happen in 2026, if Nvidia really is ahead of schedule on its latest architecture, there’s a possibility it could leap ahead and launch the Rubin architecture for consumer and gaming cards instead.

This article is from December and there's new info.
The Blackwell generation of workstation cards are being released in May: https://videocardz.com/newz/nvidia-rtx-pro-6000-blackwell-gpu-with-96gb-memory-listed-at-8435-launch-expected-in-may
Man, 96GB of VRAM would be insane for hobbyist AI stuff. I bet you could make your money back if you hopped on the porn video AI thing before anyone else did.
It seems unlikely to me that another generation of workstation stuff would release a few months later, but I admit I don't follow this business stuff. But, the way they're going to have EIGHT new workstation cards shows you their priorities when 5090s are a theoretical product. It's a broken record at this point, but nvidia obviously does not care about the people that made them successful in the first 90% of their lifespan. But, gamer cards do get released when there's enough chips that fail to pass inspection for businesses purposes the way chips that fail to be a xx90 can be placed in lower priced versions.
I wouldn't expect consumer-level 60xx generation to be released this year if that's what you're thinking. First half of 2026? I don't know about that, but who knows. I just want more VRAM.

 No.2778

I'll keep saving up money so I can afford an RTX 8080 with 256GB VRAM in 2030.




File:13696618_p0.png (197.16 KB,450x500)

 No.2728[Reply]

When it comes to IPs they always come in one of four ASN types, ones that come from ISPs, ones from businesses, ones from education, and ones from hosting. Is there any simple way of figuring out which is which in a speedy manner?

 No.2731

File:[SubsPlease] Nihon e Youko….jpg (199.58 KB,1920x1080)

I think there's a lot of them in Asia

 No.2732

File:1383252892848.png (317.97 KB,572x303)


 No.2738

File:cd676b7397de4a45b0e1152397….jpg (415.61 KB,1242x2068)


 No.2739

>>2738
Yeah that’s useful, but where do they get their data from?

 No.2751

File:eaddd01405203fc7784b9c3b15….jpg (714.88 KB,2480x3508)

>>2739
The information of the owner of ASN is provided by regional Internet registries like American Registry for Internet Numbers (ARIN) through the Registration Data Access Protocol (RDAP).
Here is the official info of AS7018 provided by ARIN through RDAP:
https://rdap.arin.net/registry/autnum/7018

This raw info only lists the owner's name, address, and contact information. But type of organization isn't a part of it. Third-party providers like ipinfo.io have their own database to determine the ASN type based on the identity of the owner.




File:[Piyoko] Himitsu no AiPri ….jpg (160.69 KB,1920x1080)

 No.2696[Reply]

Has anyone messed with local AI voice generation stuff lately? I hear there's been some advancements with stuff like https://github.com/Zyphra/Zonos/ and https://github.com/fishaudio/fish-speech but I haven't actually tried them yet. I still need to organize the raw extracted Utawarerumono voice lines for my own testing, too.

 No.2727

File:[Serenae] Kimi to Idol Pre….jpg (252.92 KB,1920x1080)

Another model is out on local: https://github.com/SesameAILabs/csm
I guess it's up to me to do some testing, so I'll see if there's a layman's guide out there.




File:1000000523.jpg (78.25 KB,563x400)

 No.2672[Reply]

I've been writing a largely custom game engine in C, C99 specifically because it's the best supported version of the language that isn't completely archaic. I'm using SDL, mostly because I want the thing to run in environments that aren't just the older version of Windows that I'm on. In doing so, I've realized two things:

First, I kind of wished I used C++. I still think C is the better overall language, and I don't exactly regret using it, but there's been a lot of situations where C++ features would've removed a lot of busywork on my end. For example, I have multiple circularly-linked lists that hold different structs but are otherwise identical in terms of access and modification, and they'd be a perfect candidate for templates. Right now, I'm handling them with a combination of boilerplate and weird post-hoc macro hacks that I put in to reduce said boilerplate when it became too much for me to manage.

Second, I've come to realize just how much of a game engine is just bookkeeping: loading data, saving data, keeping logic running at consistent speed, normalizing coordinates so the scene can be rendered, maintaining the data structures necessary for things like rendering and audio playback, etc. The problem with most of these things is that it's hard to know if they're working individually, because you need to have several of the other features up-and-running first before you get any direct feedback; I need to treat them all as one big thing, at least in the short-term, and that's a style of programming that I am not at all used to.

There is no point to this thread, by the way, I just wanted to share.
11 posts and 2 image replies omitted. Click reply to view.

 No.2695

thats very cool anon
the part that I stalled on was making a 'level loading' system because everytime i try and learn bison/lexx to make a real parser for the level files i give up

 No.2698

>>2693
>OP, is your goal to learn programming and about game engines, or is it to make a finished game?

A bit of both. I do plan to build a finished game (actually multiple!) off of this, but much of the reason I'm making my own engine is indeed so I can become a better programmer.

With that said, if push comes to shove, I'm more than willing to use libraries if it means getting the engine done in a reasonable amount of time. I'm already using SDL and I'm most certainly going to use a JSON-parsing library for resources like level-layouts and save files, and a UTF-8 library for text.

 No.2716

Quick update, a big thing I've learned working on this is the importance of documentation. I'm at the point where my code does too many things that are too complex for a quick read to suffice anymore, so I've taken to writing descriptions of everything using code comments.

 No.2720

>>2716
on topic sager

 No.2721

>>2716
Might as well take the time to create proper documentation, comments can only take you so far.




File:__sakura_mayura_and_unyuu_….jpg (298.5 KB,1444x2048)

 No.2717[Reply]

Do you use an Ukagaka (Desktop Mate)
Or have you used them?

 No.2719

I remember using a couple waaaaaay back in the day. I can't really remember much about it.




Delete Post [ ]
[Refresh] [Top] [Catalog] [Archive] [Post a Reply]

[ home / bans / all ] [ amv / jp / spg ] [ maho ] [ f / ec ] [ qa / b / poll ] [ tv / bann ] [ toggle-new ]

[1] [2] [3] [4] [5] [6] [7] [8]
| Catalog