Post

Jordan McQueen
@jm.dev

There are actually good ways to try and detect AI-generated content, but the whole “em-dashes implies generated content” approach is intellectually lazy. Hint: check for fixed-width space characters, weird Unicode hyphens, zero-width joiners, ellipsis characters, right and left quotation marks, etc.

0 reposts1 like

Jordan McQueen
@jm.dev

No, it literally uses them everywhere. You can check the output for yourself. I think they’ve added special passes at the end to purposefully insert these sentinel values to make it easier to detect generated content. Maybe it’s something they’re even doing for anti-cheating purposes.

Jordan McQueen
@jm.dev

He has taught me many things! In this case I can thank my emacs config for making it very obvious when “non-standard” characters are being used.

Jordan McQueen
@jm.dev

Yeah. Quotes is probably the weakest of all these signals. If you put U+1680 Ogham Space Mark between a number and the percentage sign, though… sketch. 😅 New game: insert weird Unicode everywhere into one’s typical usage patterns to trick people into thinking you’re a “user”. 😆