Search
OpenAI announced that starting today, API customers worldwide can opt to enable text watermarking for certain selected models. In the coming weeks, eligible text outputs from ChatGPT and Codex in the EU region will automatically include invisible watermarks. This technology, named textGrain, works by embedding invisible statistical signals during the model's word selection process. Detectors can identify these signals to determine whether the text is watermarked. Official evaluations show that its performance is on par with or superior to other solutions, though detection becomes more challenging with shorter texts or stronger wording constraints. Currently, access to the detector is limited to approved researchers and professional institutions who can apply for its use.
-
C114 Communication Network -
Communication Home
