Blog
Notes on invisible characters, model tells, and the mechanics of text governance.
Why one verb became a red flag
Model output has a vocabulary fingerprint as well as a byte fingerprint. A look at what actually gives generated prose away, and what to do about it.
ReadThe invisible characters hiding in your AI text
Zero-width spaces, bidirectional controls, and the Unicode Tags block. What they are, why they end up in generated text, and how to see them.
Read