The watermark is a relationship

See how LLM text watermarks hide weak statistical signals across many tokens.

About this project

An offline interactive lab that explains how text watermarks steer a model's word choices so a detector can recognise a pattern, with experiments on length, edits and detector keys. Source

Tools and source notes

Tools named in the source: SmolLM2-135M-Instruct.

Attribution source

Tool names are source attributions. An exact model or complete production workflow may not be disclosed.

Source: Explainer Tools

Mobile compatibility has not been confirmed.