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.
- 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.
- 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.”
- 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.
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.