Character Removal for Watermark

What problem does it solve?

This Skill addresses the challenge of detecting and disrupting hidden watermarks embedded within text generated by Large Language Models (LLMs).

Core Features & Use Cases

• Watermark Disruption: Applies character-level perturbations to text to break or weaken LLM watermarks.
• Watermark Detection Aid: Can be used in conjunction with detection tools to test the robustness of watermarking techniques.
• Use Case: A researcher wants to test if a new LLM's watermark can be bypassed. They use this Skill to modify the LLM's output and then attempt to detect the watermark on the modified text.

Quick Start

Run the character removal tool on the provided text file 'output.txt'.