SynthID or text watermarking are being looked at the wrong way if you are calling it a joke. Like the article says they are open about the limitations. It was not meant to be a universal "this is AI generated" detector.

It could be really useful for preventing synthetic data from accidentally contaminating your real datasets down the line. It's not perfect, but it's a start.

[flagged]