Jerry on PieFed

Just a techie guy running feddit.online to allow people to communicate, make friends and acquaintances. Odd coming from a happy introvert, right? (https://jerry.hear-me.blog/about)

I also own these publicly available applications:
Mastodon: https://hear-me.social/
Alternative Mastodon UI: https://phanpy.hear-me.social/
Peertube: https://my-sunshine.video/
Friendica: https://my-place.social/
Matrix: https://element.secure-channel.net/
XMPP/Jabber: https://between-us.online/
Bluesky PDS: https://blue-ocean.social/ (jerry.blue-ocean.social) Mobilizon (Facebook Events Alt): https://my-group.events/
and more…

  • 29 Posts
  • 184 Comments
Joined 1 year ago
cake
Cake day: September 29th, 2024

help-circle
  • This is definitely the best protection. If the provider drops you, you move your domain to another provider. But, as far as I know, while almost all email providers will host your personal domain, none that I know of will do it on the free plans. But your email is your identity. You should be willing to pay for it, especially if you host it on a provider that otherwise won’t make any money on you.

    There are a couple of downsides. If you forget, or are unable, to renew your domain, you lose it and your emails. Make sure another family member or friend can pay the renewal for you if, for some reason, you cannot.

    While your own domain makes it far less likely that your email will be canceled (because you can move it), abuse of your domain can result in your losing your domain name and your email, especially before it has earned a reputation.

    Which brings up another IMPORTANT point. If you use your own domain name, then you must set up your DNS records to protect your domain from spoofers and spammers so it doesn’t get blacklisted or, worse, doesn’t cause cancellation of your domain name. Scammers and spammers WILL try to send email using your domain name. You need to tell email clients to toss these rogue emails and give them the means to determine spoofing and unauthorized use. Read this: https://www.valimail.com/blog/dmarc-dkim-spf-explained/

    Also, be aware that SpamAssassin considers .com, .net, and .org TLDs to be far safer than .world, .online, .blog, and most others. Using one of these newer TLDs results in a higher spam score, and your email is more likely to end up in the spam folder if it reaches the magic score of 5. A new age TLD can add as much as 1 point to the spam calculation depending on the email provider receiving your email.

    So your own domain name is safer but costs money and requires more work.



  • I asked Gemini Pro whether the thorn character messes an LLM up? If anyone is interested, here’s the answer. The short answer is it makes it harder for humans to read the text, but LLMs are not bothered by it. So, those who are using the thorn character for “th” believing that it breaks AI are not achieving anything and are just being annoying.

    Gemini Pro’s response: (Note that as I write this in PieFed, the thorn character is being converted to “th” and so you will have to convert it back in your brain.

    The belief that using the character “þ” (Thorn) “messes with” LLM training has gained some traction in privacy-focused and “anti-AI” circles. Here is a breakdown of the technical reality behind that claim.

    1. The Theory: Disrupting Tokenization

    Large Language Models don’t read words; they read tokens (numerical representations of character clusters).

    Common Tokens: In standard English, the word “the” is one of the most common tokens. It is highly optimized, and the model “understands” it perfectly.

    The “Thorn” Hack: If you replace “the” with “<thorn>e,” you force the model’s tokenizer to handle a rare character. Because <thorn> is uncommon in English training data, the tokenizer might break “<thorn>e” into two separate, rare tokens (the byte for <thorn> and the letter “e”). Proponents believe this makes the text “invisible” or “garbage” to the model, preventing it from learning from the content or correctly categorizing it.

    1. Does it actually work? (Mostly No)

    While the idea is clever, it is generally ineffective against modern, high-scale LLMs for several reasons:

    Multilingual Training: Most modern models are trained on massive multilingual datasets. Icelandic uses <thorn> as a standard letter. Therefore, the models already have <thorn> in their vocabulary and have seen it millions of times.

    Byte-Level Tokenization: Modern tokenizers (using Byte Pair Encoding or BPE) are designed to handle any Unicode character by falling back to its raw bytes. They don’t “break” when they see an unfamiliar symbol; they just process it as a less-common sequence.

    Contextual Mapping: LLMs are excellent at pattern recognition. If a model sees a sentence like “<thorn>e cat sat on <thorn>e mat,” it can easily infer from the context that “<thorn>e” serves the exact same linguistic function as “the.”

    1. The “Glitch Token” Phenomenon

    There is a grain of truth to the “glitching” idea. **In the past, ** certain rare characters or specific strings (often called “glitch tokens”) were included in a model’s vocabulary but appeared so rarely in the training data that their “embedding” (the model’s understanding of them) was never properly trained.

    When a model encounters a true glitch token, it can sometimes “hallucinate” or produce nonsensical output. However, <thorn> is a standard UTF-8 character and is too common globally to act as a true glitch token for most general-purpose models.

    Summary

    Using <thorn> is more of a digital protest or a linguistic “speed bump” than a functional way to break an AI. It might make your text harder for a simple keyword-search bot to find, but for a modern LLM, it’s just a slightly “noisy” version of English that we can still read and understand quite easily.




  • I deleted what I wrote before. If it federated, ignore it.

    Your browser is trying to find startpage.com on your local machine instead of the Internet.

    While on the VPN, open a command window and ping startpage.com. Does it return localhost or the real IP address? If it returns the real IP address, then the problem is related to the browser. Try another browser to see if it’s Vivaldi-related.

    If it returns localhost then maybe it’s a setting in protonVPN?

    This is strange. Just try to find clues.









  • Proton stores your encrypted private key . An encrypted private key does not allow them to read your email or files.

    When you log into a new device:
    Proton sends the encrypted private key to your device.
    You type your password.
    ** Your device** (not Proton’s server) uses the password to decrypt the private key locally in your browser or app memory.

    That decrypted key is then used to decrypt your emails on your device. Proton mail sends you just the encrypted text.

    There is one potential security issue:

    Since Proton serves the website code (HTML/JavaScript) that performs the encryption, you have to trust that they serve you honest code. Proton could theoretically alter their website code to capture your password the next time you log in, which theoretically a government can force them to do.

    However, this is a different threat than “they have the keys.” Currently, they possess the keys only in a form they mathematically cannot unlock.





  • I read the post and understood the problem. Then I read Dan’s reply. And he’s fixing it. In fact, he’s fixing it in the exact way the blog post says it should be fixed, “… but those filters should be manually triggered and always removable.”

    Here’s Dan’s reply:

    “… I hear the community: you want text-only posts. We are going to build this as an opt-in feature. If you want microblogging, turn it on. If you prefer the classic media-only experience, nothing changes.”

    Why was this blog post even written then? Now it’s seeming to be personal.



  • I listened to the track. You’re really great! I see why you won! Fantastic you! BTW, I have a lot of anti-tracking software so it may not have registered in the counts.

    You mentioned depression. You and I then travel with the same evil companion. It’s good that you mentioned it. It gets weaker in the light.

    The reason to live is somewhere in the future. You can’t see it coming because it’s always around a corner; you come across it when you don’t expect it, and it’s never what you think it will be. But depression doesn’t want you to find it. It’s an evil SOB. The future will reveal why life is worth living. I know you probably don’t believe me. But, it is there.

    I’ve also known someone close who had an eating disorder. It can be dangerous. Use logic to control it. It’s also a deceitful companion.

    Anyway, you’re doing great. You have great talent, and you achieved this because you are talented and a fighter! Fantastic job! I am proud of you!