Debiasing Watermarks for Large Language Models via Maximal Coupling

Open in new window