The Linux Thread - The Autist's OS of Choice

[AnOminous]

I do suspect AI will prove to be counter-productive in a whole load of areas, including coding. It's prone not just to nonsense, but to plausible-sounding nonsense, the worst kind. It's a bullshit generator.

I certainly hope so. Because fuck the world.

 

[AmpleApricots]

something more

Absolutely. LLMs in their current state can already do things you simply cannot properly do with conventional programming. A software that can read, explain and categorize a text on a level many low level data entry kind of human workers would fail at (and in seconds, too) is no small feat and I'm not entirely sure why people dismiss that.

If LLM reason or just complete tokens in patterns at the end, well... What does a human brain do? Everything we think and do is always in the framework of things we have learned before. If you wouldn't have learned to speak english, you would not post here right now. The way in what context you learned english probably also affects what and how you post here. In a way we're all phone autocompletes. If regressive networks are the end step to AGI I don't believe either, I also don't believe there will just be suddenly a GPT69 and it'll be AGI, I feel it will be more of a gradual process with a lot of blurred lines until AI systems act in such intelligent and "reasonable" ways that the point if there really is "something" behind their camera eyes or not will be purely philosophical.

 

[Betonhaus]

If LLM reason or just complete tokens in patterns at the end, well... What does a human brain do? Everything we think and do is always in the framework of things we have learned before. If you wouldn't have learned to speak english, you would not post here right now. The way in what context you learned english probably also affects what and how you post here. In a way we're all phone autocompletes. If regressive networks are the end step to AGI I don't believe either, I also don't believe there will just be suddenly a GPT69 and it'll be AGI, I feel it will be more of a gradual process with a lot of blurred lines until AI systems act in such intelligent and "reasonable" ways that the point if there really is "something" behind their camera eyes or not will be purely philosophical

Frankly I just can't wait until they decide to replace upper management with AI, and AI determines that the best way to run their companies is to ditch the investment funds.

 

[HahaYes]

if there really is "something" behind their camera eyes or not will be purely philosophical.

Hell that's perfectly suited for concepts such as the Talos Principle, what really defines consciousness/humans etc. I won't derail the thread and sperg about it given it's about our basement dwelling OS choices, but it's one of my favorite philosophical discussions. There was a game made a while back surrounding it that's pretty good as well.

 

[Lone Wandering Courier]

As much as I consider Linux gaming to be fairly decent with exceptions (like modding), I think I'm going to switch to playing most of my PC games on Windows (atleast the proprietary ones). I've read in the past that it's a good idea to separate the environments that proprietary software and games run on and the environment that you browse sensitive stuff like emails (or you know kiwifarms) on. While it is unlikely something really bad will happen if I don't have that separation, it is probably better to be safer than sorry. Of course the added benefit is that some games will be smoother to play and mod. Another benefit is that it would be easier to justify swapping my desktop's linux distro to Artix since I won't have to worry about gaming shit that needs systemd (like the gamemode utility).

 

[Creative Username]

The sum of all written text in human history probably would fit in under 500 terabytes. The reason we talk about the Internet in terms of petabytes and exabytes is because it contains a lot of bulky information, such as pictures, music, and video. Pure text, which is what LLMs operate on, is actually quite dense.

Pardon my inaccuracy. My point still stands that it's a mind-boggling volume of data, incomprehensible in scale, that contains basically every thought that humanity has decided to write down and archive on its massive global network of interconnected information-processing machines that we refer to as the Internet.

I'm very sorry it really did sound that way but a lot really is poorly understood. To give an example for people that don't closely follow: In the last few months people in the open source sphere have figured out that a) you can basically "cut away" huge parts of a models "brain" with little effect, vastly lowering resource requirements b) you can stich several "brains" together in various ways and the resulting model sometimes is smarter than the models it's made of as a result. That's where we are at here. That's also why it really is too early to tell where it will go.

I meant that you've read more of the research and actually interacted with it more by yourself, I'm essentially a layman when it comes to this shit, and you seem to be the kind of person who can implement all of these algorithms in about a week. You're right that nobody knows how it works. You just know 0.002% about it while I'm at about 0.00135%.

As much as I consider Linux gaming to be fairly decent with exceptions (like modding), I think I'm going to switch to playing most of my PC games on Windows (atleast the proprietary ones). I've read in the past that it's a good idea to separate the environments that proprietary software and games run on and the environment that you browse sensitive stuff like emails (or you know kiwifarms) on. While it is unlikely something really bad will happen if I don't have that separation, it is probably better to be safer than sorry. Of course the added benefit is that some games will be smoother to play and mod. Another benefit is that it would be easier to justify swapping my desktop's linux distro to Artix since I won't have to worry about gaming shit that needs systemd (like the gamemode utility).

I'd like architectures like Qubes OS's to catch on more. I love the idea of running several completely incompatible computing environments alongside (and isolated) from each other and the host system at the same time.

 

[Car Won't Crank]

As much as I consider Linux gaming to be fairly decent with exceptions (like modding), I think I'm going to switch to playing most of my PC games on Windows (atleast the proprietary ones). I've read in the past that it's a good idea to separate the environments that proprietary software and games run on and the environment that you browse sensitive stuff like emails (or you know kiwifarms) on. While it is unlikely something really bad will happen if I don't have that separation, it is probably better to be safer than sorry. Of course the added benefit is that some games will be smoother to play and mod. Another benefit is that it would be easier to justify swapping my desktop's linux distro to Artix since I won't have to worry about gaming shit that needs systemd (like the gamemode utility).

Doesn't this already happen on Linux when playing windows games through wine? Each game gets its own prefix and environment. Do you mean something different?

 

[Belisarius Cawl]

Absolutely. LLMs in their current state can already do things you simply cannot properly do with conventional programming. A software that can read, explain and categorize a text on a level many low level data entry kind of human workers would fail at (and in seconds, too) is no small feat and I'm not entirely sure why people dismiss that.

I get what you're saying entirely while at the same time one can become cynical by going into full QA mode and doing things like asking about the name of H.P. Lovecraft's cat and demanding an MNIST classifier in QBASIC (which, in fairness, ChatGPT actually blocked out into subroutines out for me).

If LLM reason or just complete tokens in patterns at the end, well... What does a human brain do? Everything we think and do is always in the framework of things we have learned before. If you wouldn't have learned to speak english, you would not post here right now. The way in what context you learned english probably also affects what and how you post here. In a way we're all phone autocompletes. If regressive networks are the end step to AGI I don't believe either, I also don't believe there will just be suddenly a GPT69 and it'll be AGI, I feel it will be more of a gradual process with a lot of blurred lines until AI systems act in such intelligent and "reasonable" ways that the point if there really is "something" behind their camera eyes or not will be purely philosophical.

Sorites, my favorite paradox

 

[Lone Wandering Courier]

Doesn't this already happen on Linux when playing windows games through wine? Each game gets its own prefix and environment.

No, wine/proton has access to the whole system.

 

[SilentDuck]

I'd like architectures like Qubes OS's to catch on more. I love the idea of running several completely incompatible computing environments alongside (and isolated) from each other and the host system at the same time.

Being able to lock yourself out locally from your keyboard and mouse and stay locked out is not likely to ever catch on. I like it but would like it even more if I was a glownig on Uncle Sam's nextgen laptop budget. Dunno if Grandma wants to download, install and run a new disposable Fedora VM everytime she needs to click on the emails from her new fren in Liberia.

 

[AmpleApricots]

what really defines consciousness/humans

LLM has triggered this discussion in many places and I feel the opinions are mostly personally motivated, not really coming from a semblance of objective observation. Because it's just that hard a topic, but yes. This is a derail.

I get what you're saying entirely while at the same time one can become cynical by going into full QA mode and doing things like asking about the name of H.P. Lovecraft's cat and demanding an MNIST classifier in QBASIC (which, in fairness, ChatGPT actually blocked out into subroutines out for me).

Most people that try to get help from an LLM for their task go about it the wrong way, they approach it top-down (hey gpt, write me a program that scrapes names from websites) instead going from the bottom up (hey gpt, how would I go about downloading "example.com/index.html' with shell tools? And then "how would I search for text in the downloaded file now?" "how can I search for strings that are prefixed with the text 'Name:'?") This approach helps LLMs basically by building a context and indirectly prune possible pathways you don't want away as more and more unlikely. (since you went about building the context like this, the LLM will not try to suddenly solve the problem with python, for example) It is a useful tool if you think about the limitations. I also would not use something like the ChatGPT website directly but make calls to the API of the various models, or run a model locally. The websites have so many normie filters (which also is just a ton of context that is secretly included in every query) that they become borderline useless because the model basically gets primed for unrelated nonsense which is incredibly "distracting" for an LLM.

separate the environments

My brother in Linux, have you considered namespaces? Linux let's you isolate programs in various ways, e.g. network namespaces you can put a program in so it cannot communicate with other programs outside the namespace or over your network interfaces. You can do this on a very granular, file system level where a program can e.g. only access certain files in /etc/, hell even only certain devices in /dev and /sys. They are an immensely powerful tool to isolate without having to use an VM. I mostly use bwrap, you can write scripts with that really easy.

Here is a simple skeleton script as an example that uses bwrap. You can put it in a directory, then you create two subdirectories ("home" and "game"). "game" contains your game files, "home" will be the folder that contains dotfiles, for wine for example, so you won't clutter your filesystem and can have a custom enviroment for each wine-ran game. The program started inside bwrap will not have access to your network or home folder. This script assumes a very simple system using X and ALSA with a global ALSA config. You might have to allow local connections to your X server, depending on your setup.

#!/bin/sh

CMD="wine [add name of game exe here]"

bwrap   \
                    --unshare-all \
                    --dir /home/user \
                    --bind home /home/user \
                    --bind game /home/user/game \
[you might have to adjust the following lines for your distro, no commenting here, remove this line]
                    --ro-bind /usr /usr \
                    --ro-bind /bin /bin \
                    --ro-bind /lib /lib \
                    --ro-bind /etc /etc \
                    --dev /dev \
                    --proc /proc \
                    --dev-bind /dev/dri /dev/dri \
                    --dev-bind /dev/snd /dev/snd \
                    --perms 1755 --dir /sys \
                    --ro-bind /sys /sys \
                    --ro-bind /tmp/.X11-unix /tmp/.X11-unix \
                    --clearenv \
                    --setenv DISPLAY "$DISPLAY" \
                    --setenv LC_ALL C \
                    --setenv WINEDEBUG -all \
                    --setenv HOME /home/user \
                    --die-with-parent \
                    --new-session \
                    --chdir /home/user/game \
                    $CMD

This is very lax and basically just covers the bases that wine doesn't have more enviroment variables than it needs to, that files in your home directory can't be read/written to (outside of the game and home folder you just created) that the program cannot escalate privileges, and that the program cannot access the internet, which usually is more than enough to diminish the danger of most programs. (try out to see what's visible by changing CMD to your shell interpreter) It also turns wine's debugging messages off which really can slow down programs. This script can be adjusted and improved upon in various ways. You could for example start the game in a different X server without session and window management, which increases security and also helps with lag caused by e.g. compositors. You could change the first block of ro-binds to point at the installation of a different distribution, this comes in handy if you want to run glibc binaries in a musl distribution, for example. You could also carefully analyze which absolute minimum of files a program needs with ldd (be careful with ldd if you don't trust the program, it's basically like executing it) and strace and tighten it up a bit further. This script allows access to all gpus and sound cards of the system via dev, and to absolutely everything (eg temperature sensors and battery status - SuperIO) via /sys. You could improve on it further to only give the minimum needed access to the gpu and nothing else via sys, e.g.:

--ro-bind "/sys/devices/$GPUPATH" "/sys/devices/$GPUPATH"
--ro-bind /sys/dev/char /sys/dev/char \

where GPUPATH is the PCI path of the GPU, e.g.
GPUPATH="pci0000:00/0000:00:02.0"

This is also handy when you have several GPUs in your system and want to make sure a specific one is picked.

Sky's the limit with bwrap, it gives you very granual control.

You can also run programs in different network namespaces and have e.g. start all programs without network access by default and only move them to the appropiate network namespace if you want them to have network access.

 

[jONALD]

The current iteration of stupid AI, or at least the one I use is a fun toy and a bit of a nice tool for creative writing (not just coomer slop). It can write some rather nice prose with enough tweaking of settings and a few knee slapping one liners from time to time. Forget writing code, I want the current stupid ones uncensored for fun writing assistants like Claude 3.

 

[Rusty Metal Skull Gun]

My brother in Linux, have you considered namespaces?

This is really cool - I'd not heard of bwrap before. One weakness of this particular script is that if you have RW access to /home, you can put whatever malicious thing you want in ~/.local/share/Autostart, or whatever it's called nowadays.

But why bother? Just buy games from reputable stores like Steam or GOG, if you're worried about malware.

 

[Heliantheae]

On Qubes security https://seclists.org/dailydave/2010/q3/29.

 

[AmpleApricots]

RW access to /home

that's the fun of it, the RW access to home is to the "container'home in (whatever directory the script is in)home/, not the real home. The real $HOME is completely invisible to whatever program bwrap started. (and it's children) If you run that script and set CMD as /bin/bash, then enter "ls $HOME" you will see that it is empty. If you enter "touch $HOME/somefile" the "somefile" will be in (whatever directory the script is in)home/.

And if you leave "--bind home /home/user \" out, the home directory will only exist in RAM for the duration of the program that bwrap called and also only be accessible to this program. Try it.

The structure is as follows (lets assume the game is called "daggerfall")

daggerfall/
├── bwrap-script.sh
├── game (containing your game files)
└── home (for dot files the program that is started via bwrap-script creates, savegames etc.)

/home/rusty/daggerfall/home will be mounted as /home/user, /home/rusty/daggerfall/game will be mounted as /home/user/game. The daggerfall executable will never see or be able to interact with your real home directory. For all purposes, your real home directory doesn't exist inside this container.

And if you want to get rid of the whole thing:
rm -r daggerfall

simple as.

EDIT: fixed paths

 

[Betonhaus]

https://github.com/ifupdown-ng/ifupdown-ng/issues/234

People have found that ifupdown-ng may have similar social engineering attacks, and it is part of Debian.

 

[Ahriman]

https://discourse.ubuntu.com/t/xz-liblzma-security-update-post-2/43801/3?u=d0od

 

[Heliantheae]

https://github.com/ifupdown-ng/ifupdown-ng/issues/234

People have found that ifupdown-ng may have similar social engineering attacks, and it is part of Debian.

That's very interesting. That Neustradamus might be a useful idiot judging by his linked socials, he's been doing his thing for quite a while.. Which is the baffling thing to me; what sort of existence is announcing new versions of myriad of software, and pushing for using new versions of libs for software, that you cannot possibly be using all at once? Why? What does he gain by this? Why does he have followers interested in this? Is it some sort of paraphilia for bleeding edge software?

 

[⠠⠠⠅⠑⠋⠋⠁⠇⠎ ⠠⠠⠊⠎ ⠠⠠⠁ ⠠⠠⠋⠁⠛]

Why? What does he gain by this? Why does he have followers interested in this? Is it some sort of paraphilia for bleeding edge software?

It's called 'Arch user disease'.

 

[The Ghost of Kviv]

This is also handy when you have several GPUs in your system and want to make sure a specific one is picked.

There's another way to select a specific render device with Vulkan I learned recently. Mesa provides an environment variable called MESA_VK_DEVICE_SELECT. For example, to see a list of render devices:

MESA_VK_DEVICE_SELECT=list vulkaninfo