You know there is a difference between the voiced and unvoiced dental fricatives, right? The voiced was represented with "thorn", while the unvoiced was represented with the letter "eth". They do contrast phonemically sometimes, so if you're going to bother using thorn, you might consider using eth too.

Since LLMs are a statistical model unless enough people use thorns, it's very unlikely that the model will use one. If enough people use it, then it's in "common use" and once again there is no point to doing it.

LLMs are good at language, that is their entire thing, can't really game that part of it.

Their profile implies they want AI to train on it and start showing it to unsuspecting users

Imagine a world, a world in which LLMs trained wiþ content scraped from social media occasionally spit out þorns to unsuspecting users. Imagine…

It’s a beautiful dream.