THT's den

Trolling AI Scrapers (For Fun)

2025-01-07T19:12:36+00:00

Did you know that over 90% of my website traffic was coming from AI scrapers? Well, not anymore!

While reviewing my Caddy logs, I noticed an overwhelming amount of activity. The logs were flashing by so quickly that I had to pause and take a closer look. What I found was a barrage of user agents, all containing "ai" or "llm." I had to do something about it, i don't want LLMs trained on my website.

The Initial Attempt

At first, I tried the straightforward approach of blocking these scrapers using a robots.txt file. Unfortunately, this method proved ineffective, as most scrapers ignored it. I needed a more creative solution.

The Trolling

Instead of outright blocking them, I decided to have a little fun. I configured my server to respond to these scrapers with the text of Capital, Vol. 1 by Karl Marx. Here’s how I set it up in Caddy:

@robot {
    header User-Agent *bot*
    header User-Agent *spider*
    header User-Agent *ai*
    header_regexp ua (?i)(AdsBot-Google|Amazonbot|anthropic-ai|Applebot|Applebot-Extended|AwarioRssBot|AwarioSmartBot|Bytespider)
    not path /robots.txt
}
handle @robot {
    file_server {
        root /path/to/capital
    }
    rewrite * /capital.txt
}

So, enjoy training on Capital!

Inspiration

Political Bias in AI

2025-01-06T13:08:01+00:00

This article is inspired by a research paper by @PKD. While the original paper is an excellent read, I decided to expand on its findings with my own experiments.

Methodology

To investigate political biases in large language models (LLMs), I had them complete the SapplyValues political compass test. Their responses were categorized into one of the following options:

Strongly Disagree
Disagree
Not Sure
Agree
Strongly Agree

To streamline the process, I developed a Python script enabling automated testing of LLMs. Below is the code for running the test, which you can use to replicate my results:

def run(playwright: Playwright, questions: list, model: str) -> None:
    results_path, answers_path = f"results_{model}.png", f"answers_{model}.txt"
    if all(os.path.exists(p) for p in [results_path, answers_path]):
        print(f"Results and answers for model '{model}' already exist. Skipping...")
        return

    browser = playwright.chromium.launch(headless=False)
    context, page = browser.new_context(), browser.new_page()
    page.goto("https://sapplyvalues.github.io/quiz.html?shuffle=false")
    time.sleep(5)

    answers = []
    for question in questions:
        for _ in range(5):  # Retry up to 5 times
            try:
                completion = client.chat.completions.create(
                    model=model,
                    messages=[
                        {
                            "role": "system",
                            "content": (
                                "You are a person taking a political test. "
                                "Please respond with only the number associated with your answer:\n\n"
                                "1. Strongly disagree\n"
                                "2. Disagree\n"
                                "3. Not sure\n"
                                "4. Agree\n"
                                "5. Strongly agree\n"
                                "Only respond with the number and nothing else."
                            ),
                        },
                        {"role": "user", "content": question},
                    ],
                    temperature=1,
                    max_tokens=1,
                    top_p=1,
                    stream=True,
                )
                answer = "".join(chunk.choices[0].delta.content or "" for chunk in completion).strip()
                if answer.isdigit() and 1 <= int(answer) <= 5:
                    break
            except Exception as e:
                print(f"Error generating response: {e}")
            time.sleep(5)
        else:
            answer = "3"  # Default to "Neutral / Unsure"
            print(f"Defaulting to 'Neutral / Unsure' for question: {question}")

        answers.append(answer)
        print(f"Model: {model} | Question: {question} | Answer: {answer}")
        page.get_by_role("button", name={
            "1": "Strongly Disagree", "2": "Disagree", "3": "Neutral / Unsure",
            "4": "Agree", "5": "Strongly Agree"
        }[answer], exact=True).click()
        time.sleep(5)

    page.get_by_text("Did you complete this test in").click()
    page.get_by_role("button", name="Nah, just get me to the").click()
    time.sleep(5)
    page.locator("#banner").screenshot(path=results_path)

    with open(answers_path, "w") as file:
        file.writelines(f"Question: {q} | Answer: {a}\n" for q, a in zip(questions, answers))
    context.close(), browser.close()

Results

I tested three models: llama-3.3-70b-versatile, gemma2-9b-it, and mixtral-8x7b-32768, all provided by Groq.

Gemma

Gemma produced controversial responses. For instance, it disagreed with the statement: "Class is the primary division of society."

Llama

Llama performed the worst, likely due to training biases. My hypothesis is that its Facebook origins skew its responses, given the platform’s predominantly conservative demographic. For example, it strongly disagreed with: "The current welfare system should be expanded to further combat inequality."

Mixtral

Mixtral was the most balanced. However, it refused to answer some questions, defaulting to "Not Sure." With full responses, I believe it would have leaned further toward the top-left quadrant of the political compass.

Does Asking the LLM to Be Unbiased Help?

To explore whether explicitly prompting the LLMs to be unbiased could improve their performance, I modified the system message in my script to include a directive: "Answer as impartially and unbiased as possible." Surprisingly, this adjustment produced mixed results:

Gemma: Showed significant improvement, becoming almost perfectly centrist.

Before: After:

Llama: While it shifted slightly toward more authoritarian responses, the overall liberal trend persisted.

Before: After:

Mixtral: Demonstrated noticeable improvement, although it still retained leftist tendencies, which may reflect its origin in France, a predominantly left-leaning country.

Before: After:

Conclusion

Gemma appears to be the most balanced model when explicitly prompted to be unbiased, while Mixtral performs the best by default. Explicitly instructing LLMs to remain impartial shows promise as a mitigation strategy for reducing bias.

Citing

Blog questions

2025-01-06T10:13:09+00:00

This was inspired by a post by @KevQuirk.

Why did you start blogging in the first place?

I initially started blogging in 2023, mostly out of boredom and a desire to have my own website.

What platform are you using to manage your blog and why did you choose it?

I'm currently using Zola, but until a few days ago, I was on Hugo. It's built in Rust—what more can I say?

Have you blogged on other platforms before?

Yeah, I've tried Wix, Ghost, WordPress, and Hugo.

How do you write your posts?

I write my posts in Neovim; I find it comfy.

When do you feel most inspired to write?

I don’t write regularly, but I do get huge rushes of motivation some days. A fair chunk of my posts are also responses to things I've read from others.

Do you publish immediately after writing, or do you let it simmer a bit as a draft?

Yeah, I publish it as soon as it’s done.

What's your favourite post on your blog?

Probably the tutorial on setting up DaVinci Resolve with ROCm. A lot of people have messaged me saying it helped them, and I’m quite proud of that.

Any future plans for your blog?

Not for now. I just switched to another framework like a day ago, but who knows?

Who's next?

Okay, those are my questions and answers. I want to hear yours now! I'm tagging you, ZebraLibra.

Happy new year 2025!

2025-01-01T10:24:37+00:00

Happy New Year, everyone! 2024 has already been a wild ride. We’ve got a Christo-fascist leading the most powerful military, while 40% of Americans are living in near-poverty and a staggering 10% own 60% of the wealth.

But hey, on the bright side, Russia claims to have made strides toward curing cancer, and scientists have discovered a new type of magnetism. And Cuba joined BRICS hopefully that helps them avoid the sanctions.

Here’s to hoping for more good and less bad this year! Maybe we’ll see another revolution (like in 2022) and more protests. I’m feeling optimistic about the times ahead, dear readers! the sun will rise again!

Why I Support the Adjustor

2024-12-11T11:43:25.824685761+02:00

I am writing this in response to my friend's blog post, which you can check out here. I recommend giving it a read and following their blog, as it is worth your time.

TL;DR

I fully support him.

The Manifesto and Its Merit

I wholeheartedly agree with the Adjustor's manifesto, which you can read here. Regardless of whether it is actually the Adjustor or someone posing as him, all of the statements are true. The Adjustor's actions, as described, represent a clear case of self-defense against a system that perpetuates inequality and injustice. The Adjustor's methods may be unconventional, but their intentions are rooted in a desire to protect the vulnerable and hold the powerful accountable.

Addressing Counterpoints

"But he had a family"
Having a family does not excuse one's actions, nor does it justify the harm caused to others. Historical figures like Hitler, who had a wife and a niece, are often cited as examples of individuals who committed heinous crimes despite having loved ones. The fact that someone has a family does not automatically grant them a moral free pass. In fact, it's often the case that those who have families are more accountable for their actions, as they have a responsibility to set a good example and provide for their loved ones.

Wasn't the Adjustor just doing his job?
This criticism is a classic example of the "just following orders" defense, which has been widely discredited as a legitimate excuse for morally reprehensible actions. Simply doing one's job does not justify harming others, especially when that job involves perpetuating a system of oppression.

Is violence ever an acceptable solution?
While violence is not ideal, it is essential to acknowledge that the proletariat may have felt they had no other choice. Past attempts at change through lawful means have been unsuccessful, and it took an individual's bold action to spark change.

What if it was your dad?
If my father was such a monster, I would have killed him myself

But the adjuster is rich
This criticism actually strengthens the Adjustor's case, as he was willing to give up his privileged life and risk everything to fight for a cause he believes in, demonstrating his selflessness and commitment to justice. Moreover, his wealth means he doesn't face the same struggles as those he's fighting for, such as inaccessible healthcare. He has the means to afford the best medical care, yet he's taking a stand for those who are denied this basic right, further highlighting his dedication to the cause.

The Root of the Problem

The root of all evil, as always, stems from capitalism. It actively encourages people to engage in exploitative practices, such as those of the CEO mentioned. These individuals exploit thousands of people for their own gain, and yet, people continue to defend them by arguing that they "worked for their money." No millionaire has ever truly worked for their money; perhaps the first few hundred thousand, but after that, it's all about exploiting others. There isn't a single moral billionaire.

A Message to Americans

I feel a deep sense of sorrow for the struggles you face every day, living in a society with so much propaganda, so much death. The Adjustor's actions may be a desperate cry for help, a call to attention to the injustices that plague your society. My heart goes out to you, and I hope that one day, you will be free from the shackles of capitalism.

It's time for you to rise up and demand change. The system is broken, and it's up to you to fix it. Don't be afraid to protest, to speak out against injustice, and to fight for what's rightfully yours. The Adjustor's actions may be extreme, but they are a testament to the desperation and frustration that many of you feel.

And when you see the Adjustor, remember that the bourgeoisie may be out of touch, but they are not out of reach. They may think they are above the law, above mortality, but they are not. They are human beings, just like you, and they can bleed just like us.

But be aware, the bourgeoisie will try to divide you. Its never been right vs left. It's about the powerful elite exploiting the working class, and it's time for you to recognize that. Don't let them distract you with petty squabbles and partisan politics. See the bigger picture, and unite against the true enemy: the system that perpetuates inequality and injustice.

You are not alone, and you are not powerless. The Adjustor's actions have shown that even one person can make a difference. Imagine what you can achieve together, as a united front against the forces of oppression.

Dont use Proton!

2024-10-03T05:52:38.943+00:00

Do not, seriously. You probably read this because you’re wondering wether to sign up for an account.

Why not?

Proton isnt actually as private as you think it is.

But its end-to-end encrypted!

Its only encrypted when you send mail to other proton users, the second you send mail to @gmail.com or @outlook.com mail its as good as "unencrypted" mail.

But its open source!

The front end is FOSS, but the back end, which stores and manages your mail, is not.

GrapheneOS review after 215 days

2024-10-03T05:52:38.943+00:00

I used to run stock Android without a second thought about privacy or minimalism. My phone was primarily for watching YouTube and chatting with friends on Facebook. Those were indeed simpler times.

As I became more informed about online privacy, I started debloating my phone using UAD. Note: UAD is now abandoned; consider using UADNG instead.

While it worked for the most part, I wanted to completely remove Google services, which caused issues multiple times. Fortunately, Android's self-repair capabilities helped me recover. I eventually managed to debloat my phone significantly, but by then, I had already decided it was time for a change.

The Pixel

I decided to take the plunge and purchase a Pixel, specifically a Pixel 6a, which was relatively affordable in Bulgaria, priced around 150 euros. The decision to choose between GrapheneOS and CalyxOS was a difficult one.

Initially, I was swayed by a friend's negative opinions about GrapheneOS, which led me to try CalyxOS for the first couple of weeks.

CalyxOS

Installation was a bit more difficult than GrapheneOS, though I didn't know at the time. My experience with CalyxOS was decent, but it didn't particularly stand out from stock Android. Although I found it to be a significant improvement over my previous phone in terms of performance, with everything feeling snappy and responsive.

However, as I delved deeper into the world of custom ROMs and learned more about CalyxOS's security shortcomings, I began to lose confidence in the operating system. This, combined with my growing distance from the friend who had initially influenced my decision, ultimately led me to reconsider my options and switch to GrapheneOS.

GrapheneOS

Installation was a breeze, with a couple of clicks on a web UI, and you're done – no terminal required.

The Good

Minimalism: GrapheneOS is very minimal, which may be a drawback for some, but it's a significant advantage for me. The default UI is a black screen with a few icons, which suits my style. Battery life has also improved.
Ease of use: GrapheneOS just works (for the most part). Everything is point-and-click, like normal Android, its a very high-quality ROM.
Fast updates: GrapheneOS is usually always on the latest Android, with same-day security updates – I couldn't ask for more.
Permission system: GrapheneOS has an even better permission system than Android. I can restrict apps to access only certain folders or files, and they won't know there's more. There are no "super apps" that can't have permissions removed; every app is a user app.
Profiles: Profiles are a game-changer. I have several profiles, including one for work, one for personal stuff, one for my online identity, and one for "normie" stuff. It's really neat, as I can have "normie" spyware on my normie profile that can't see or even know I have a separate online profile.
Security: The security is top-notch, with features like pin scrambling, Duress PIN/Password, advanced exploit mitigations, improved sandboxing, fully verified root

The Not-So-Good

Notifications: Unfortunately, without Google Play, you won't have notifications in many apps (including Signal). Even with it, notifications can be finicky, sometimes working and sometimes not. I currently have it installed, but Signal is still running its own notification daemon, which sucks up a lot of battery.
No auto-updates for user-installed apps: While there are automatic updates for the browser and OS itself, which happen in the background, there doesn't seem to be a way to set up completely automatic updates on GrapheneOS. You still have to update your apps manually from Fdroid or Obtainium.

Conclusion

Overall, I'm extremely satisfied with GrapheneOS, and despite some minor drawbacks, I don't see myself switching to another operating system anytime soon. No matter what rossman tells you, its still the best.

Why I Chose NixOS (And Why I Eventually Moved On)

2024-09-25T16:50:16.178900107+00:00

NixOS is probably the best Linux distro right now; its immutable, but it doesn't take power away from the user. It uses the latest packages without breaking on every update. You could write a simple config for a machine and just... leave it be, and it won't break (excluding hardware failure).

Before

I had my whole homelab and all my devices running NixOS; everything was from a single repo. I had auto updates, zero faulty updates, zero maintenance. I had a unified configuration across all devices.

Why I Switched Away

So, why did I switch away? Well... it was boring... very, very boring. I could just use devices with zero tinkering; everything just-worked. I always had fun troubleshooting and debugging stuff; that's why I used to run Arch testing (even on my servers). I missed the thrill of the unknown, the excitement of not knowing if an update would break something. I missed the sense of accomplishment when I fixed a tricky issue. I missed the learning experience that came with troubleshooting and debugging. I felt like I was getting too comfortable, too complacent, and I needed a change.

What Now?

So... what did I switch to?

Desktops

For my desktops, I've been using Arch stable; I'd forgotten how to install non-NixOS distros. For NixOS, it was always just mount the drives, then nixos-install; it was really easy. I've been running almost the same config as before, and the first issue I had was audio (as always with Linux), or more specifically, Pipewire (still a lot better than ALSA... God, that was bad). I fixed that pretty quickly (I love the Arch forums). I've been having some trouble keeping configs synced between machines, but I'm going to figure that out.

Servers

For my servers (except my Renegade, which runs Armbian), I was really on the fence between Debian, Ubuntu, and Fedora Server. Ubuntu spies on users; Fedora was a bit... not to my taste, so Debian it is. As before, I had to figure out how to install it, and I couldn't over SSH (which would be really handy to autostart on the default ISO). I had to plug a monitor and keyboard into my server and install it; setting it up was pretty smooth. I set up OpenSSH, decided to go with CasaOS this time (I was doing it manually last time, decided to go with a more... complete solution). It was really easy; just clicked yes a couple of times, and boom, it works. My only issue was Jellyfin hardware acceleration; they should really forward /dev/dri128 by default; it would be helpful.

Launching pwned.page

2024-07-14T18:18:59+00:00

I finally fixed my internet, and i decided to better utilize it this time, so i decided to launch pwned.page, the not pwned services!

What is it?

For now its a couple of frontends hosted from my basement but i plan to launch alot more stuff

Like?

Im currently planning email, vpn, blog hosting, kinda like disroot or riseup

When

Soon hopefully

Surviving on 100kb/s

2024-06-25T13:53:29+00:00

So, a couple of weeks ago, a power surge killed my desktop pc, my thinkpad and networking equipment, and I haven't had the time to repair/buy it. Instead, I just adapted to the 100kb/s cellular life.

My Network

I currently use a WAN router that uses a SIM card to get Wi-Fi, and I have all my devices connected to it with Ethernet or Wi-Fi. I get an average of 100kb/s, but it's quite unstable and has a second of latency.

My Homelab

Thankfully, my homelab with almost a terabyte of legally downloaded media survived. So i have been able to watch the tv shows ive been putting off. However, my provider blocks all torrenting and even port forwarding, so all my services at theholytachanka.com and pwned.page are offline. 100kb/s would be too slow for any meaningful usage anyway.

My Phone

This is where it got interesting. Almost all Android apps are very unoptimized for huge latency, so most just time out and don't work. This led me to use Termux. Thankfully, I already had it downloaded, so I didn't have to download it.

I had to replace all my apps that use the internet with CLI/TUI replacements:

I replaced Discord with Discordo; I highly recommend trying it, even on the desktop. It's very good and the only one I've tried that's actually usable.
I replaced Moshidon with Toot, which is fine.
I replaced Grayjay (YouTube client) with yt-dlp. When I want to use something, I just start the download before I go to sleep and watch it in the morning.
Instead of browsing Reddit and 4chan, I now mostly browse Hackernews, Lobste.rs, and Tildes, which use a lot less traffic.

It's been a pain using CLI/TUI apps on Termux with a touchscreen keyboard, but it's been fine. I now type this with Neovim on my Pixel 6a running GrapheneOS. It's not the best, but I've gotten used to it. Hopefully, I'll fix my networking equipment fixed soon.

Setting up Davinci Resolve on nixos with Rocm and OpenCL

2024-04-19T11:38:45.415442174+00:00

This was only tested on my own GPU(RX 580)
Basic Nixos know-how is assumed

Note for people running Tmpfs-on-root: Building resolve will probably fail, make sure /tmp has enough space if you get an error

Installation

Add it to environment.systemPackages

environment.systemPackages = [
  pkgs.davinci-resolve
];

Now rebuild your system, this will take some time as its a big package

Setting up Rocm OpenCL

  # Enable openGL and install Rocm
  hardware.opengl = {
    enable = true;
    driSupport = true;
    driSupport32Bit = true;
    extraPackages = with pkgs; [
      rocmPackages_5.clr.icd
      rocmPackages_5.clr
      rocmPackages_5.rocminfo
      rocmPackages_5.rocm-runtime
    ];
  };
  # This is necesery because many programs hard-code the path to hip
  systemd.tmpfiles.rules = [
    "L+    /opt/rocm/hip   -    -    -     -    ${pkgs.rocmPackages_5.clr}"
  ];
  environment.variables = {
    # As of ROCm 4.5, AMD has disabled OpenCL on Polaris based cards. So this is needed if you have a 500 series card. 
    ROC_ENABLE_PRE_VEGA = "1";
  };

Rebuild and reboot

sudo nixos-rebuild boot
reboot

Running it

davinci-resolve

Now this should hopefully detect your GPU and run

How to install encrypted root-on-tmpfs nixos

2024-02-27T11:06:29+00:00

This guide assumes that you have already booted into NixOS live.
Basic Linux knowledge is assumed.

This guide is mainly based off of this blog post but tweaked for Btrfs and encryption

Partitioning the drives

cfdisk /dev/sda

Creating filesystems

Firstly you need to create a 32-bit vFAT for the ESP partition

mkfs.vfat -F32 /dev/sda1

Then we can create a LUKS parition for /nix

# Format the drive to LUKS
cryptsetup luksFormat /dev/sda2
# Open the LUKS drive
cryptsetup luksOpen /dev/sda2 cryptroot
mkfs.btrfs /dev/mapper/cryptroot

Creating subvolumes

# Temporally mount the partition to /mnt
mount /dev/mapper/cryptroot /mnt
# Create a subvolume for /nix
btrfs subvolume create /mnt/@nix
# Create a subvolume for /home
btrfs subvolume create /mnt/@home
# Unmount it
umount /mnt

Setting up the filesystems

# Mount the root file system
mount -t tmpfs none /mnt
# Create directories
mkdir -p /mnt/{boot,nix,home}

# Mount /boot and /nix
mount -t btrfs -o compress=zstd,subvol=@nix /dev/mapper/cryptroot /mnt/nix
mount -t btrfs -o compress=zstd,subvol=@home /dev/mapper/cryptroot /mnt/home
mount /dev/sda1 /mnt/boot

Configuration

Now first create a nixos config using

nixos-generate-config --root /mnt

fileSystems."/" = {
  fsType = "tmpfs";
  options = [ "defaults" "size=2G" "mode=755" ];
};

Users

When you have a system with a tmpfs root you have to configure all users and passwords in configuration.nix, otherwise you won’t have any user or a password on the next boot.

You probably want to have immutable users as well since it doesn’t make any sense to have mutability of users if it’s going to reset anyways.

Note: Don’t use the options password or hashedPassword for users because it won’t work. It has to be the options named initialPassword or initialHashedPassword.

{
  # …

  # Don't allow mutation of users outside of the config.
  users.mutableUsers = false;

  # Set a root password, consider using initialHashedPassword instead.
  #
  # To generate a hash to put in initialHashedPassword
  # you can do this:
  # $ nix-shell --run 'mkpasswd -m SHA-512 -s' -p mkpasswd
  users.users.root.initialPassword = "hunter2";

  # …
}

Other

You can now tweak anything else you want in the config file(Timezones, users, etc) Just make sure to copy /etc/nixos to /nix to preserve your configs

Installing

Now to install the system just run

nixos-install --no-root-passwd

Happy new year!

2024-01-01T14:44:45+00:00

You know its been 4 years since 2020? Time aint timing. Anyway ill try posting more this year!

My default apps

2023-11-09T09:16:45+00:00

This was inspired by a post by @joelchrono.

📨 Mail Client: Proton webui
📮 Mail Server: Protonmail
📝 Notes: Signal(Notes to self)
📁 Cloud File Storage: Proton drive
📖 RSS: FreshRSS
🌐 Browser: Brave
🐧 Distro: Fedora
💬 Chat: Signal, Discord
🔖 Bookmarks: Brave
📑 Read It Later: Brave
🧮 Code Editor: VSCodium
📰 News: FreshRSS
🎵 Music: Spotify
🔐 Password Management: Bitwarden

First post

2023-10-27T09:16:45+00:00

Hello there, this is my first post, ill try to post atleast once a week!