Update of 8/21/2026: How Claude’s text watermark works "... we, along with several other major AI providers, are implementing this change to comply with the EU AI Act. ...
Claude’s text watermark is a version of the SynthID-Text approach published by Google DeepMind in a Nature paper in 2024. It belongs to a family of approaches that go back to a proposal by Scott Aaronson in 2022, all of which share the same design principle that we described above—the watermark only changes the source of the randomness used to pick among words. ..."
The term watermarking in the digital age seems to be an oxymoron and antiquated itself! It sounds very cute, familiar too many and may convey reliability and peace of mind to some.
However, when it comes to e.g. text content is such watermarking feasible and robust? I have serious doubts. Apparently one simple technique seems to be to insert unusual/unexpected terms into a sequence of regular/usual terms.
With other media/content like photos, images or video this might be more feasible.
Or are some tech companies just primarily trying to please some ridiculous EU regulations? You bet!
No comments:
Post a Comment