Large-scale online deanonymization with LLMs

https://lemmus.org/post/20438053

I have an account where I only post after I translated my writing through three different languages and back to English. The original input and the output convey the same message, but have very distinct styles. Randomizing the three languages in my translation sequence introduces enough variety that I doubt current LLM’s can identify me. (Full disclosure: I don’t post any sensitive information under any account; I do it just for fun.)