[yorwba]

Section 6, "Removing Watermarks," of the paper you cite makes it very clear that detecting LLM-generated text is not solved if the user takes measures to avoid detection.