2026-08-19 08:02 UTC
Requiring invisible watermarks instead of a visible “made with AI” label is baffling.
If the goal is to tell readers when text is generated by AI, a clear “made with AI” tag is immediately obvious, no extra tools needed.
By contrast, an invisible statistical mark only helps if readers can already detect it. Detection may be available as a separate API, but pricing and access aren’t public yet. Will it be free and open, or gated behind accounts?
From my perspective, even open-weight models running on personal hardware seem to fall within scope. Does that align with your reading?
I’m genuinely curious how others interpret this. Am I missing something?
#AIWatermark #EUAIAct #AIRegulation #AITransparency #llms #softwareDevelopment #softwareEngineering #machineLearning #eu #digital #digitalMarketing
Replies (1)
-
@JMkinen@mementomori.social 2026-08-19 09:33
@sevepy@social.vivaldi.net Wouldn't that kind of text label be just removed by the user/bot most of the time before the text is read by anyone else? With statistical fingerprinting it might be tiny bit trickier.