Many commenters (and the paper) are thinking about the watermarking in adversarial settings, e.g. detecting students using AI assistance improperly.

But I think even simple watermarking probably has value; consider a corporate context in which the corporation itself may want to monitor and know what proportion of the code, content, or work product is AI-generated. In that setting, fairly simple markers would allow at least a rough estimate or indication, although they'd have the converse problem of not necessarily indicating places where humans did some hand-editing of the output.