This is true and the author seems to not understand the problems with greedy (top 1) decoding or the fact that watermarking affects only high entropy tokens. But the published watermarking methods still have a slight negative effect on perplexity, so there is something more to it.