Watermark Yourself

Moderators sound like moderators

Recently there was a big hullaballoo about the fact that Claude is watermarking their generated text. John Gruber called it a 'perversion of writing':

My error was believing Anthropic that their system wouldn’t adulterate and corrupt the semantics of the text their models generate. That is in fact exactly what they plan to do.

— Daring Fireball, John Gruber

Anthropic themselves claim that the watermark is imperceptible and that you won't be able to detect it.

When a supported Claude model generates text, it weaves an imperceptible watermark directly into the text itself. You won’t see it, and it doesn’t change the meaning, quality, or readability of Claude’s response.

— Claude Support

Back in the day, there was a website that would identify people's writing style and suggest other users who write just like them. There aren't that many HN users with over 2000 words and 50 comments and these days experimenting with something like that is fairly easy so I thought I'd give it a shot.

The idea is quite trivial: take 2/3/4-char-grams across all users that have 2000 words, and 50 comments, strip out quoted-text and code-blocks (people use either for quotes), and weight them according to tf–idf. Then I just precomputed the results for everyone and stuck them in a SQLite database and you can now query HN users by style for fun.