Vertically stacked thanks to the use of #GNUScreen on my host system, three #Linux setups running simultaneously, competing for lightweightness. (The software you see running inside all of them is the htop)

All systems are headless, and two of them are really, really headless, meaning no video adapter and limited to interfacing through serial port ttyS0 (although I could also use SSH, but I chose not to).

The bottom-most system is #ArchLinux (btw), a distro I've been long familiar with. It's set up with #VirtualBox VMSVGA Graphics adapter, so it's not exactly headless.

The middle one is #AlpineLinux, a distro often used for Docker containers. It really has no Graphics adapter and, still, it managed to boot gracefully. I'm interfacing with it through its ttyS0, and I can even see the Plymouth Boot Screen through the serial port (because I configured GRUB to use serial along with console).

The top-most system is #VoidLinux, a distro whose ISO can't be booted with less than 2GB of RAM (and trust me, I tried, and I faced a Kernel Panic every time I tried to reduce RAM for its Live ISO). I managed to install it with a setting that allows for way less RAM. Similarly to Alpine, it also has no Graphics adapter and no SSH daemon, so I'm interfacing with it through serial.

All the three systems have 256 MB of "physical" (VM) RAM, but the total memory (MemTotal) is less due to certain memory reservations. Alpine and Arch both have 203MB of MemTotal (53MB less than the physical due to kernel and other things stealing from total memory), while Void has 197MB of MemTotal (its kernel code, kernel bss, kernel data and kernel rodata sums up to 5MB additionally needed mem reservation, and SATA controller is using a little more mem reservation than PIIX4).

Void's root fs is using 2.1GB from a 7.3GB partition (within a 8GB virtual disk).
Alpine's root fs is using 213MB from a 3GB partition (within a 4GB virtual disk).
Arch's root fs is using 4.6GB (to be fair, there are some media files there) from a 13.2GB partition (within a 16GB disk)

As you can see, Arch is heavy to this task. Don't take me wrong: Arch is an excellent distro. It's either good for desktop (be it bare-metal or virtualized), for gaming and for bare-metal servers. But when it comes for a headless server, it got too much weight to carry from its own, as you can see by its memory usage. systemd-journald alone is using 8.7% of mem, while NetworkManager is using 9.3%. For comparison, Alpine's syslogd, also responsible for daemon logging, is using just 0.4%.

It's just a casual test. My purpose is to find the lightweightest VM with "routing" capabilities: there's a fourth VM (an Arch with LXQT desktop) which needs to be isolated from my host's LAN (i.e., no routing to my own host, and no routing to the other devices within my LAN) while being able to access the Internet (and that's why I'm not willing to have a SSH daemon, although I did set up such service within the Arch VM). The aforementioned Arch VM is currently the one doing this job, configured with iptables, although the screenshot was taken in a scenario where I didn't boot the fourth VM, just the trio Void, Alpine and headless Arch.

It's interesting to see a system gracefully booting without any screen, yet getting to offer 256-color terminal through serial interface.

I got the farthest I could to make my lightest VM ever, totally headless (i.e., no video at all) while still keeping network-related functionalities:

- It's an UEFI Void Linux setup, whose live ISO originally can't boot with less than 2GB (tried to boot with less than 2GB before installation just to face a kernel panic). I installed it, then I started to find out how far I could reduce its RAM.
- No video at all, no SSH either: I configured both the GRUB and the Linux to use the serial port. I can access it using `socat` and `screen`/`minicom` together. I can choose which device to boot at GRUB (although there's no need for choosing other than the Void Linux that's installed there), I can see the plymouth boot "screen" (without a screen), and I can login and have a full `bash` shell (in the screennshot, I'm using `dash` to hide the hostname).
- There's an iptables inside, as well as two interfaces, because I'm intending to set it up as a tiny router to isolate another VM from the LAN while allowing it (the other VM) to access the WAN (the Internet).

Purpose: not that I needed to make it lighter (than it already was before removing
the video)... however, I did it "because it can be done".

Some issues:
- I couldn't get to remove `btrfs`: although I instructed `dracut` to omit the `btrfs` module, as well as blacklisting it through `/etc/modprobe.d/blacklist.conf`, it's still being loaded somehow, as the plymouth shows " Activating btrfs devices... Scanning for Btrfs filesystems". There's no btrfs partition within the setup, just the swap, ext4 and fat32 (vfat) (the EFI partition).
- I still need to find a way to `stty cols .. rows ..` automatically (i.e., having host's `screen` and guest's `getty`/`bash`/whatever to communicate somehow regarding host `screen`'s current terminal size), outside a SSH environment. I avoided configuring a SSH daemon just because it can be done.
- VirtualBox isn't made for truly-headless VMs. While one can set "Graphics Adapter" to "None", setting 0MB of graphical memory supposedly makes it glitch.

Screenshot:

LLMs are known for requiring lots of electricity to function, and this is mostly due to how AI corps aren't (weren't?) required to make things as efficient as possible (because they're backed by essentially "infinite money cheat"). Then DeepSeek unexpectedly appeared, bringing a thing that is insanely optimized and efficient (just compare how many GPUs they needed versus how many GPUs OpenAI et al used). And since they openly released the weights, there are many sizes available as of now: from those running on their servers, all the way to models capable of being run in a modest smartphone. You read it right: these things can run without needing power-hungry GPUs, literally in the palm of the hand. It's definitely way less harm to environment, since a smartphone cannot pull more energy than it's physically capable of pulling from a USB cable (and if it's unplugged, it'll necessarily rely on the battery).

So I tried to run a distilled version of DeepSeek locally in my smartphone (my phone is not a flagship). It's not as perfect as a big model (the ones running on GPUs), but it's... Interesting. Notice the wattmeter in the notification bar: it's pulling almost a single ampere (1A × 5V = 5W) while it's "thinking". The fact I can run a damn LLM using just 5W and a mobile device is just mesmerizing.

On the one hand, when Deep Seek "doesn't know" about a thing (i.e., something not present the training data), it'll state it clearly (I'm not sure if the image will be sent as I'm not using Lemmy directly to reply this):

The context of the image is the following: I asked DeepSeek about "Abnukta", an obscure and not-so-much-known Enochian term that is used during one of the invocations of Lilith, and DeepSeek replied the following:

"Abnukta is a term that does not have a widely recognized or established meaning in mainstream English dictionaries or common usage. It could potentially be a misspelling, a neologism, or a term from a specific dialect, jargon, or cultural context. If you have more context or details about where you encountered the term, I might be able to provide a more accurate explanation. Alternatively, it could be a name or a term from a specific field or community that is not widely known".

So, the answer that the user Rimu received is not regarding something "unknown" to the LLM (otherwise it'd be clearly stated as that, as per my example), but something that triggered moderation mechanisms. So, in a sense, yes, the LLM refused to answer...

However... On the other hand, western LLMs are full of "safeguards" (shouldn't we call these as censorship, too?) regarding certain themes, so it's not an exclusivity of Chinese LLMs. For example:
- I can't talk about demonolatry (the worshiping of daemonic entities, as present in my own personal beliefs) with Claude, it'll ask me to choose another subject.
- I can't talk with Bing Copilot about some of my own goth drawings.
- Specifically regarding socio-economics-politics subjects, people can't talk with ChatGPT and Google Gemini about a certain person involved in a recent US event, whose name is the same as a video-game character known for wearing a green hat and being the brother of another character that enters pipes and seeks to set free a princess.
- GitHub Copilot refuses (in a blatant Scumthorpe Problem) to reply or suggest completion for code containing terms such as "trans" or "gender" (it's an open and known issue on GitHub, so far unanswered why or how to make Copilot answer).

But yeah, west is the land of the freedom /s

Some experimentation with Chaos Game.

For those unfamiliar with Chaos Game, it's a mathematical experimentation (not really a "game") which applies some constraints to randomness, so order and pattern will emerge from it.

For example: you have an equilateral triangle and you start at the topmost vertex of it. You randomly choose between the three vertices, then you place a point exactly in the middle between the current coordinates and the chosen vertex. You repeat the process indefinitely, choosing another vertex and placing a point midway between the current coordinates (it's now a place inside the triangle) and the chosen vertex. Gradually, you'll start to notice a pattern, in this exact case, you'll end up with a Sierpinski triangle.

The more random and unbiased the randomness, the more quality the end result will have.

You can change parameters. For example, you can change the geometrical shape (e.g. a pentagon instead of a triangle), or you can change the ratio (i.e. how far from the current coordinate to the next coordinate: instead of going half the way, you can choose to go 1/3 (0.3333...) of the way, or you can even extrapolate past the next coordinate with a ratio bigger than 1). You can choose to not repeat the previous coordinate, or you can choose to not choose a neighboring vertex, and so on... Different constraints will result in different results, sometimes a fuzzy cloud of dots, sometimes a pattern.

It's interesting to see patterns emerging from randomness.

(I tried to send two pictures, but one of them got an error while I tried to upload...)

@alexadeswift While I'm not exactly Wiccan (I'm closer to a Luciferian and Lilithian perspective/belief, with many syncretic aspects), I see Death as one of the aspects/archetypes from The Dark Mother Goddess (Lilith), known as Agrat Bat Mahlat, who would be akin to the Crone aspect of Hekate.

My understanding of Goddess Lilith (and consequently, of Na'amah and Agrat Bat Mahlat) is grounded both on (unexpected) individual/personal spiritual experience with Her, as well as on a Brazilian book, "Encontro com Lilith: mulher, demônio e Deusa" (Encounter with Lilith: woman, demoness and Goddess) written by Mark H. Williams, which has an entire section on how to experience Lilith, including what he calls "Caminhada no cemitério" (Walk in the cemetery):

English translation:

Walk in the cemetery

One of the primary Lilith roles is the transformation force. Although these transformations occur in different ways, facing Death is a key step to our path. If Death is a hard subject for you, perhaps it's better to postpone this practice until you've already worked with Lilith for a while.

Find a nearby cemetery, where you feel comfortable to walk alone. Before entering, throw some coins as an offering to the spirits of guarding the cemetery. Ask permission and thank them for their services to the living and the dead.

Start at one point in the cemetery and go through it all, imagining Lilith by your side. Observe some gravestones: notice the names, birth/death dates, among other details. If a grave catches your eye, linger over it for a while, meditating about the person who's buuried there and what would their life have been like. Ask Lilith to help the soul of the deceased as well as helping you in your journey, too.

I did this several times. Strikingly, when I first time I practiced this, I hadn't the book yet, I was still discovering and trying to figure out about Her on my own. It was shockingly surprising to me when I bought the book (because the title having Lilith's name caught my eye at the bookstore) and was faced by the exact practice I did before.

As I said before, from a syncretic perspective, Lilith could be seen as Hekate with Her three supreme aspects: Crone (Agrat), Mother (Lilith Herself) and Maiden (Na'amah). The very figure of The Death is sometimes seen as a Goddess, such as Santa Muerte and Morana.

I hope this helps!