OPENAI
The EU told AI to wear a name tag
OpenAI is adding invisible watermarks to text made by ChatGPT and Codex in the EU to meet new AI transparency rules.
The watermark will roll out over the coming weeks to eligible users on all plans. It will not appear as a visible label.
Instead, OpenAI’s system, called textGrain, slightly changes word choices to create a hidden pattern that detectors can spot.
Here’s what’s happening:
Where: ChatGPT and Codex text in the EU
How: Hidden patterns in word choices
Limit: Editing can make detection less reliable
The thesaurus strikes back
OpenAI says the watermark does not identify users or noticeably affect model performance.
However, it can become harder to detect if the text is edited, translated or very short.
In one test, changing 10% of the words cut detection from about 92% to 66%.
For now, OpenAI’s detector will only be available to approved researchers and expert groups.
The company also says that no watermark does not prove a text was written by a human.
Ctrl+C, Ctrl+V, Ctrl+Crime Scene. -MG