To comply with the EU's new AI transparency rules, OpenAI will add an invisible watermark to text and code generated in the EU. Should AI-written text always be labeled? This site names the model behind every machine translation; how much disclosure is enough?

I wish that were possible, but I think it's realistically difficult. Even a simple model running locally could get around that watermark, and the more AI develops, the harder it will get.

I had this site disclose the translation model so that readers can gauge the possibility and scope of translation errors, and also so that the translation quality can be improved in bulk later on. It was an engineering decision, not an extension of that broader discussion.