Pseudpocalypse

According to a new LessWrong essay, pseudonymous blogging faces an emerging threat from de-anonymization through writing style analysis. Author dynomight argues that writing patterns leave enough of a statistical fingerprint to link different pseudonyms—even from small samples. Current evidence supports this: large language models like Claude can already identify an author from roughly one thousand words of text. Using information theory, dynomight argues that about twenty-nine bits of distinctive writing information would be sufficient to connect pseudonyms among four hundred ninety million English speakers. The essay extends further, suggesting the same principle applies to physical identifiers—gait, body shape, scrolling patterns. Dynomight expects this de-anonymization to accelerate as AI improves, potentially making anonymity untenable across most high-bandwidth interaction.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton