[ home / rules / faq / search ] [ overboard / sfw / alt ] [ leftypol / edu / labor / siberia / lgbt / latam / hobby / tech / games / anime / music / draw / AKM ] [ meta ] [ wiki / shop / tv / tiktok / twitter / patreon ] [ GET / ref / marx / booru ]

/tech/ - Technology

"Technology reveals the active relation of man to nature" - Karl Marx
Name
Options
Subject
Comment
Flag
File
Embed
Password(For file deletion.)
What is 6 - 2?
Check out our new store at shop.leftypol.org!
| Catalog | Home
|

File: 1760866353466.jpg (216.06 KB, 800x800, 6850faba3197480093f6b22347….jpg)

 

they're making "AI GPUs" now. it's not a GPU at that point it's an AIPU. we don't call CPUs "central GPUs".

they should call them AIPUs and market them as that.
1 post omitted.

>they should call them AIPUs and market them as that.
They are though??? It's right there in your picture. Huawei called it an inference card.

Have heard the term LPU, language processing unit from Groq.
And NPU used by Microsoft.
Seems there's just not a standard yet.
Language seems over restrictive, NPU sounds closer.

>>31597
AIPU too clunky to say, IPU/inference card or NPU will prolly win in the long term


Sorry kiddo those babies pump out 4 bit calculations at Petaflops, if you want to use it for vidya start looking at making your own motherboard inspired by the Sony Playstation 2

https://venturebeat.com/ai/nvidia-researchers-unlock-4-bit-llm-training-that-matches-8-bit-performance

>>31599
APU was unfortunately taken by AMD's marketing team pretending that an integrated GPU was novel or useful in a new way.

My money is on NPU.


File: 1755139966457.png (8.38 KB, 389x129, ClipboardImage.png)

 

The other thread hit bump limit and I'm addicted to talking about the birth of the ̶a̶l̶l̶-̶k̶n̶o̶w̶i̶n̶g̶ ̶c̶o̶m̶p̶u̶t̶e̶r̶ ̶g̶o̶d̶ the biggest financial bubble in history and the coming jobless eschaton, post your AI news here

Previous thread: >>27559
299 posts and 48 image replies omitted.

>>31572
I mean at that point you just want an autistic kid that knows VIM

ai video gen economics are nonsense if you think about them, sora 2 generates a video for a user, said user reposts it on tiktok, insta and xitter, monetizes it, and gets a marginal cut. hosting costs are pennis on the dollars for a social network, but generation price is like 5 dollars at least per video (in the unlikely event the video was created in one shot, most likely it's significantly more expensive because the user kept cranking the lever to get something acceptable), so in essence openAI is subsidizing meta and tiktok, no wonder theyre so invested in making their own social network. so in the near future they're not going to let users download AI generated content, it'll have to be posted in the same place it was generated, because social media is technically funneling money away from AI providers.

>>31585
No way inference is that expensive unless your amortizing training costs into it.

<OpenAI, the company behind ChatGPT, released their own browser called Atlas, and it actually is something new: the first browser that actively fights against the web.
https://www.anildash.com//2025/10/22/atlas-anti-web-browser/
This is a lovely article.
<When I first got Atlas up and running, I tried giving it the easiest and most obvious tasks I could possibly give it. I looked up "Taylor Swift showgirl" to see if it would give me links to videos or playlists to watch or listen to the most popular music on the charts right now; this has to be just about the easiest possible prompt.

<The results that came back looked like a web page, but they weren't. Instead, what I got was something closer to a last-minute book report written by a kid who had mostly plagiarized Wikipedia. The response mentioned some basic biographical information and had a few photos. Now we know that AI tools are prone to this kind of confabulation, but this is new, because it felt like I was in a web browser, typing into a search box on the Internet. And here's what was most notable: there was no link to her website.
And it gets worse from there.

Musk launches Grokipedia! Hooray!
<My Grokipedia entry has over seven thousand words, compared to a mere 1,300 in my Wikipedia article. It’s pretty clear how it was generated; an LLM, trained on who-knows-what but definitely including that Wikipedia article and this blog, was told to go nuts.
Doesn't that sound amazing?
<Every paragraph contains significant errors.
https://www.tbray.org/ongoing/When/202x/2025/10/28/Grokipedia


File: 1732711575153-0.mp4 (21.61 MB, 576x1024, 45900c8dc0a663193a0b8d04a1….mp4)

File: 1732711575153-1.png (37.89 KB, 264x264, 1731568844360.png)

 

<Exposure to AI output immediately damages your brain's cognitive abilities
>9/10th's or more of 4chan, reddit and twitter activity is GPT-J bots
>Most video platforms curate suggestions using AI
>Search engines place AI output at top of results
>several platforms implement AI summaries
>Language translation programs being replaced with AI
>Browsers and closed source operating systems integrating AI features into UX
>Schools using AI to mark unusual vocabulary as suspect of being AI generated, advantaging students with average or below literacy.

I know the AI bubble is crashing so this won't continue to be a problem in 5 years, but it is worrying that we don't know how long this brain damage lasts.
50 posts and 8 image replies omitted.

>>27264
Isn't "office work" most prone to bullshit jobs in the first place. Ripe not for automation by AI, but for erasure.

>>31217
Well offices are a property value thing, make buildings where hundreds of people do shit that could've easily been done from home by one gal with linux on her laptop that knows bash script, and you have a commercial building, with all the zoning laws and tax implications that entails. And that's profitable enough that you can pay someone to be the "senior stapler inspector" or "blockchain synergy consultant."

>>27264
but it doesn't automate the office jobs though. because it gets shit wrong a lot. so in reality, at best it's used as a tool by a human being who has to double check all of it's work to make sure it's correct. and office work isn't like work on an assembly line where in theory there is an unlimited amount value you could produce were you not limited by your own body. in any given day there is a finite number of human to human communications tasks that have to be done by an office worker (and are the core of most of their jobs) which put a hard cap on the value they produce in a day and then they just play minesweeper or whatever and do their best to look busy

>>31099
I've started making regular trips into the woods near my house, down a natural stream because it's the lowest part and creates a natural trail through what is essentially untouched forest and it's extremely nice to be out there and just center myself. There's deep water at some points so I've taken to just hiking in completely barefoot for about 2km.

It's become my new favorite hobby, you're not bored, you're not anxious, you're not worried about what people think, you're not in society, you don't have technology, any negative emotion evaporates about 1-2 hours in. The worst part of the entire thing is the walk back to my house. If it wasn't switching to winter I think I would just bring a tent out there and start living out there.

b-boobs


File: 1758091596451.png (9.29 KB, 500x143, Briar_SVG_Logo.svg.png)

 

What do you guys, gals and enbies think about Briar?
It's a p2p communication program but for once has a different approach.

From wiki:
>…communications with no centralized servers and minimal reliance on external infrastructure. Messages can be transmitted through Bluetooth, Wi-Fi, over the internet via Tor or removable storage, such as USB sticks.

Anyway, seems like a neat idea but maybe I'm missing something

I used to use Briar for a while. it's pretty neat. for those who don't know, its main selling point is that it is metadata resistant. its main downside when I used it was battery usage, because it had to stay connected to Tor. message delivery is also an issue, because both users have to be online. in a group chat this is easier, but it's still necessary for a critical mass of people to be online for messages to percolate

Boomp


File: 1756509712830.webp (2.33 MB, 2528x1346, asjg9sgjs.webp)

 

Is a self-driving car really an invention that society needs? We already have a machine that was fundamentally designed for a human operator, there is no entity in the universe more well-suited to the task of driving an automobile than a human being, but now we want to try to retrofit every car with some software that replaces the human driver and will never be as good as a human driver and for what? Is driving really such a laborious and horrible ordeal for humans to endure, having to sit in a comfortable air-conditioned chair and listen to music for a while? Is it really worth wiping out an entire category of jobs that humans depend on to survive, just to replace them with software that is not nearly as well suited to the task and will likely kill people with no accountability?
7 posts omitted.

There's no point to anything

>Is a self-driving car really an invention that society needs
Probably, because of points that have already been brought up. Personally I'm wary of them because if they become ubiquitous they might try to take away human-drivable cars. As someone who enjoys driving I'm not interested in that.

>>31606
Massively gay

>>31610
>massively gay
Like you? Anyways it doesn't matter be as gay as you want

Are self-driving buses gay

no. the personal motorized carriage, self driving or not, as a mode of transport and the subsequent design of infrastructure, industry and institutions around it have been a disaster for the human race.


File: 1761479465853.png (1.32 MB, 1280x720, ClipboardImage.png)

 

Do you think it's worth it to look for decent secondhand hardware as windows 10 is EOL?

I like my crusty ol' thinkpad, but I want something more modern (read: quieter and more powerful). Suggestions are welcome.

>suggestions
NixOS with niri window manager, with KDE window manager in case niri's sattelite-based window positioning borks something (i.e. Krita when using a tablet pen)
Basically install NixOS with KDE then add niri.
Niri is good for general use as it's lightweight.

>>31608
I meant hardware doofus

The EOL is not a sharp cliff.

1. You can still get updates for free with a Microsoft account.
2. Most businesses have already gotten rid of their pre-11 inventory. If you're looking for ebay deals this is where a lot of them come from.
3. A lot of people, especially sellers, install Windows 11 on incompatible hardware, even hardware that's too old to receive updates because the ISA is too outdated.

So you won't see the kind of glut in the market that you might think. That said you've been able to get a cheap T470 for years if that's what you want. It is probably cheaper than it would be if it was officially compatible with Windows 11.

>>31607
windows 10 iot ltsc is still an option. that or linux

I think windows 11 iot ltsc bypasses hardware requirements anyways so it doesnt matter


File: 1761154341724.png (459.65 KB, 1008x720, 1384993837.png)

 

I'm writing a tech demo of an idea I had about alternative GUI programming techniques. so far I had been using xcb and everything is going fine and the idea is working but I was thinking about adding wayland support before releasing the code. I know wayland has a compatibility layer with x11 but considering I'm not doing anything crazy, mostly just drawing rectangles and text, and handling events, I thought "how hard could it be to add native support with a few ifdefs here and there". however, it doesn't seem like there is much documentation on how to write applications for wayland, and most places just tell you to use gtk or kde or some other high-level framework

tl;dr my question is, is there something like this official xcb tutorial but for whatever the wayland equivalent of xcb is?
https://www.x.org/releases/X11R7.7-RC1/doc/libxcb/tutorial/index.html

Looks like maybe XDG-shell and libwayland-client. [^1][^2][^3]

:[^1] https://bugaevc.gitbooks.io/writing-wayland-clients/content/beyond-the-black-square/drawing.html
:[^2] https://bugaevc.gitbooks.io/writing-wayland-clients/content/beyond-the-black-square/xdg-shell.html
:[^3] https://wayland-book.com/xdg-shell-basics/example-code.html

>>31601
huh, apparently it doesn't have drawing commands like x11, you just render client-side. I guess I will just use cairo then
but man it feels like a regression, as antiquated as it was, I liked that, in theory, you could use pure x11 programs through a network and the remote application would just send you the drawing commands to be rasterized locally

>>31602
It is pretty good, even locally if you're running a browser in a chroot or container like myself.

It's because wayland is garbage, put it in the bin.


File: 1728030622672.jpg (105 KB, 820x1024, 53y3soh1e3981.jpg)

 

(Copypasted from a previous 4chin /g/ thread as a foundation to making these generals on leftypol)
Users of all levels are welcome to ask questions about GNU/Linux and share their experiences.

* Please be civil, notice the "Friendly" in every Friendly GNU/Linux Thread *

Before asking for help, please check our list of resources.

If you would like to try out GNU/Linux you can do one of the following:
0) Install a GNU/Linux distribution of your choice in a Virtual Machine.
1) Use a live image and to boot directly into the GNU/Linux distribution without installing anything.
2) Dual boot the GNU/Linux distribution of your choice along with Windows or macOS.
3) Go balls deep and replace everything with GNU/Linux.

Resources: Please spend at least a minute to check a web search engine with your question.
*Many free software projects have active mailing lists.
Post too long. Click here to view the full text.
196 posts and 17 image replies omitted.

>>31405
Use snap install and do a lsblk, then stab out your eyes

So, i tried installing CachyOS today. Live CD looked like a standard "Arch for lazy and/or stupid people" kind of thing… At first glance. First problem i noticed, was that with a 2-and-something GB image here was no offline install option. Yeah, weird, but it's 2025, so i could theoretically forgive that. Well, it was actually a problem for me, since government is doing some weird shit with the internet, a lot of mirrors were inaccessible, but that's not the point. So, i was going to set up some VPN or proxy, but then… Something happened. Something truely, minblowingly terrible.
Don't open the spoilers less you lose your faith in humanity:
No, seriously, this is a real war crime, if you open next spoiler you're going to lose that last little shred of faith in Linux community you didn't even know you had.
"bash: man: command not found"
MAN, MOTHER OF GOD FUCKING HELL, MAN
Yeah, we've seen shitty distros, but THIS… I never would have imagine anyone could sink this low.
Curse upon anyone responsible for this abomination.

Am considering getting a pinetab2 8gb for tutoring work.
The idea is to use a patched build of chromium for the mpp. [^1]
This should alleviate the CPU pressure from recording the small built in.
It should be able to run an unrecorded 1080p separate monitor for references too.
My concern is running the un-GPU-accelerated whiteboard.
Maybe krita could work, which has partial GPU-acceleration?
But this would seem to be a little less a convenient usage of the platform.
And I'm not sure if it would work anyway…
Anyone with experience on a RK3566 think it could handle it?

[^1]: https://github.com/JeffyCN/libv4l-rkmpp

>>31595
>The idea is to use a patched build of chromium for the mpp. [^1]
>This should alleviate the CPU pressure from recording the small built in.
There's apparently not really negotiation on the encoding.
It's entirely possible that wyzant doesn't support the hardware encoding.

>>31596
>The idea is to use a patched build of chromium for the mpp. [^1]
Hardware accelerated encode is a (non-ARM) ChromeOS, Windows, or Mac OS exclusive.
Was looking into "brunch" as a way around this, maybe with crostini.
And meanwhile learned that ChromeOS is being replaced by Android…

The cleanest solution found was to use the ChromeOS kernel in Gentoo.
Use this Gentoo box to host LXC containers running the ChromeOS ROOT-A.

After all this it's possible that hardware acceleration still wouldn't work with the tutoring platform.
This is because of "negotiations" on the WebRTC encoding.
But probably it would work…


File: 1760630271687.pdf (8.23 MB, 57x255, Dr. Joseph Slavit - The Ne….pdf)

 

I'm a volunteer for Marxists.org. Finding forgotten gold articles from 100 years ago and sharing them with the modern world is my jam. The problem is that they're often microfilm scans that are a pain in the ass to read, so I have to transcribe them - which rarely goes smoothly with my OCR software. A lot of the time I have to resort to typing everything out by sight, which as you can imagine takes forever.

That OCR software is ABBYY FineReader 15, said to be the best when I pirated it right before the big machine learning breakthroughs. Is "AI" able to work magic for optical character recognition now?

Attached is a book-length article I'd like to transcribe. It's mostly too fuzzy for FineReader 15 to handle. I was originally going to call on /leftypol/ to help me transcribe it by hand, but I thought I'd ask /tech/ first to see if a machine can do it after all.

TL;DR: help me transcribe this plz.
12 posts and 1 image reply omitted.

>>31586
Better yet do encode image by image, but not in a stupid way:

def render_pages(
    src_pdf: pathlib.Path,
    resolution: int,
) -> List[bytes]:
    pages = []
    with Image() as images:
        images.resolution=resolution
        images.read(filename=src_pdf)
        images.depth=8
        images.colorspace="gray"
        for image in images.sequence:
            compressed = Image(image=image)
            compressed.format = 'jpeg'
            buffer = io.BytesIO()
            compressed.save(file=buffer)
            pages.append(buffer.getvalue())
    return pages

>>31586
I'll admit that I have no idea how to use that code or what it means, but thanks for creating the script to transcribe this! There are a fair amount of errors, but not as many as ABBYY would generate - it seems like the LLM is a lot better at cutting out all the extraneous exponents and apostrophes that ABBYY would pick up from the film grain.

The downside from the LLM is that you sometimes get some true bizarre hallucinations. Like this one in the opening sentence of part IV. Here it is typed out by sight:

In the first article we introduced the reader to Comrade William English Walling, the "new" Duehring, who proposes a "new" Socialism based on "new" methods and principles.

And this is what the LLM spat out:

In the first article we introduced the reader to Conrad Williams*. [*Footnote: Not William, as given in the heading. Editors.] Dorothy, who proposes a "new" Socialism based on "new" methods and principles.

To clarify, there are no footnotes. The AI just totally made that up somehow.

>>31590
>no idea how to use that code or what it means
Well, that's a little too bad. I'm not exactly sure how to run it on Windows either. But it sounds like your existing solution is maybe good enough?

>true bizarre hallucinations
Yah, one of my other runs (maybe the one with the weaker model?) got it repeating "to my knowledge" a hundred times and monotonically decreasing the column width until it was just a word or part of one for one article.

>>31591
>But it sounds like your existing solution is maybe good enough?
I have dozens of hours of experience using it, which helps a lot. ABBYY also has a lot of PDF image editing tools built-in to deskew the text etc., so it looks like I'd be processing through ABBYY anyway before running the images through Tessaract or an LLM. Severely grainy or unevenly exposed images still give it a lot of trouble through with random apostrophes etc. that are a chore to manually remove.

I thank you again for transcribing the Slavit text for me though. It looks good enough that manually correcting it wouldn't be too bad. We'll see about that though, maybe the hallucinations will be so severe I'd have to verify every sentence.

The ultimate solution to this problem will be to get New York University to take the original crinkly newspapers out of storage and scan the broadsheets properly. Which the publication definitely deserves, but it's a huge ask and I'd like to be able finish my William English Walling complete works project (plus polemics aimed at him) before I dive into that.

>>31592
Good luck with your endeavors then.

>William English Walling
This seems like a very interesting character and project.


File: 1751762900436.png (3.11 KB, 225x225, images.png)

 

>WireGuard
Written in C, so no.
>OpenVPN
Insecure (and written in C).
>Tailscore
Proprietary trash. Again, written in a lower level language.

When the fuck will somebody use Python to script a VPN protocol out of for a full tunnel client w/ access to iptables w/ default setting at "strict"? Plus run it on custom STUN servers, uses 10.8.0.53 instead of 8.8.8.8 (Google)/1.1.1.1 Cloudflare)/9.9.9.9 (Quad9) and blocks any connections to them for a fine-grained DNS control, and fully self-hosted w/ local only control. Everything else is too insecure. Oh, and also
>Tailscale leverages Google's OAuth2 for user authentication, allowing users to log in to Tailscale using their Google accounts
Lmao. The absolute state.
16 posts omitted.

This is in OCaml plus a unikernel so free of bloat: https://blog.robur.coop/articles/miragevpn-server.html

Nebula[1]?

[1]: https://github.com/slackhq/nebula

what's wrong with mullvad

https://github.com/sshuttle/sshuttle

If you want to go with the proxy client + VPS route, Singbox, Clash or V2ray are written in Go


https://github.com/SagerNet/sing-box
https://github.com/MetaCubeX/mihomo
https://github.com/v2ray/v2ray-core


Delete Post [ ]
[ home / rules / faq / search ] [ overboard / sfw / alt ] [ leftypol / edu / labor / siberia / lgbt / latam / hobby / tech / games / anime / music / draw / AKM ] [ meta ] [ wiki / shop / tv / tiktok / twitter / patreon ] [ GET / ref / marx / booru ]
Previous[ 1 /2 /3 /4 /5 /6 /7 /8 /9 /10 /11 /12 /13 /14 /15 /16 /17 /18 /19 /20 /21 /22 /23 /24 /25 /26 /27 /28 /29 /30 /31 /32 /33 /34 /35 /36 ]
| Catalog | Home