OpenAI announced it will introduce text watermarking for content generated by ChatGPT and Codex to comply with EU transparency regulations.
1 editorial report · 3 verified social mentions. The most authoritative report leads while later evidence completes the story.
🤖 Our approach to EU text provenance rules How OpenAI is approaching text watermarking under EU rules. Learn where watermarks apply, how detection works, and why access starts with researchers. 📰 Source: OpenAI News 🔗 Link: https://openai.com/index/eu-text-provenance # AI # ArtificialIntelligence
Open mentionOpenAI just announced that they will start to apply watermarking on their model’s output text and images to comply with the EU regulation that wants to be able to identify whether a text or an image was produced by an AI model. Anthropic announced the same thing a while ago (they applied it worldwide, not just in EU). The way they do that as they explained is by enforcing a “statistical signal in text generated”, meaning preferring not always the most appropriate next token but close enough, in order to meet the “statistical signal” requirement. In my understanding, this deteriorates the output quality of their models, as it introduces KLD>0. And we know that any KLD divergence greater than 0 (which the watermarking certainly creates) may be negligible in small outputs, but it definitely b
Open mentionOpenAI is adding text watermarking in ChatGPT and Codex
Open mention