Skip to content

Comment on Follow-Up Thoughts on Watermarking Schemes for AI-Generated Text

Comments

I still fail to see what exactly is the issue that is being raised beside "watermarking bad"

Watermarking is a new tool, it's useful. Are you a website aggregator with human review? watermark the content to reduce the workload.

Are you automatically tagging novels in a creative writing forum? check for the watermark.

Are you presented with extraordinary evidence for which the provenance is extremely relevant? look for the watermark and that will tell you if the evidence is fabricated

Has someone written an amazing scienific result and is claiming ownership? watermark tells how much the attribution should go to the human or the machine.

There is nothing "wrong" or no "getting caught" in those scenarios, those are just extra tools we can use in a world where text can be arbitrarily generated. This is no different than a "made in china/designed in california" tag, or a CE marking on your product.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.